Eve library

OSHUN Unified Product Gap Analysis, Nisaba/Sophia Integration, Admin Surface Strategy, and Backing Systems Audit

Guide

Revision 1

On this page

Date: March 22, 2026
Author: Codex
Scope: apps/oshun/web, apps/oshun/mobile, apps/oshun/bff, libs/oshun/*, libs/nisaba/*, apps/sophia/*, libs/sophia/*, libs/openapi/*, libs/proto/*, libs/contracts/iris/*, libs/contracts/psyche/*, docs/domains/isis/*, docs/integrations/civitai/*, apps/yemaya/*, apps/lilith/*, services/psyche/*, apps/psyche/admin/*, libs/aphrodite/research-agents/*, openapi/isis/openapi.yaml, and the existing Oshun product/domain/integration documentation


1. Executive Summary#

OSHUN already has the beginnings of a unified shell, but it is not yet a truly unified product.

The repo shows a clear shell-first ambition:

  • one account
  • one navigation model
  • one home/explore/activity/profile shell
  • one shared domain registry
  • one cross-domain recommendation/search/activity direction

But the implementation reality on March 22, 2026 is still closer to "four domains inside one shell" than "one coherent product with multiple expressions of a shared philosophy."

The biggest findings are:

  1. The shell still hard-codes four domains almost everywhere. tara, veritas, nyx, and arete are treated as a type-level invariant across navigation, registry, auth, analytics, notifications, assistant, search, and BFF helpers. nisaba exists in the repo as a substantial library ecosystem and documentation set, but it is not exposed by the Oshun web app or mobile app.
  2. Tara is positioned as a domain, but not yet as the product's governing theme. If Tara, meditation, awakening, and enlightenment are meant to be the main theme, then Tara should function as Oshun's experiential spine: start-of-day centering, transitions, reflection, sleep, and meaning-making. Right now Tara is one strong domain among peers, not the shell's orienting principle.
  3. Web and mobile are not at the same maturity layer. The web app is domain-deep, with many per-domain routes and richer content surfaces. The mobile app is shell-rich but still domain-shallow, with most domain depth compressed into a single generic route and domain-specific mock/demo content.
  4. Nisaba is strategically the highest-leverage next domain, but only if it is integrated as a contextual knowledge layer, not just a fifth icon. The repo already contains meaningful Nisaba capabilities under libs/nisaba/*: research workspace, reference resolution, comparative texts, manuscripts, annotations, collation, translations, morphology, lexicon, timeline, and influence-network primitives. The app shell does not yet expose any of this.
  5. "Full integration" should not mean equal surface area on every platform. It should mean coherent role clarity. Tara and Arete need strong mobile-native habit loops. Veritas needs fast cross-platform briefings and research depth. Nyx needs event-driven ritual and sky tooling. Nisaba should be web-deep and mobile-companion-first. Cohesion matters more than identical parity.
  6. Sophia already exists as a substantial research and grounding domain, but it should not be added naively as another consumer card. The repo positions it as the knowledge backbone: ingestion, semantic search, citation workflows, knowledge graph, evaluation, and a workbench UI. The right move is to make Sophia the shared grounding substrate behind Nisaba, Veritas, assistant, and studio experiences.
  7. The agentic and media-generation stack is ambitious but unevenly productionized. Isis has the clearest control-plane posture for generation. Sophia has the clearest grounding posture for research and citations. But parts of Yemaya, Lilith, and Aphrodite still rely on in-memory runtime state, stub fallbacks, or simulation-style implementations. Those patterns are acceptable for prototyping but not as the invisible substrate for a trust-sensitive, quality-sensitive unified product.
  8. Oshun also lacks a unified operator product. The repo contains meaningful admin and reviewer surfaces in Psyche Admin, Sophia Workbench, Oshun Studio, Lilith review components, and BFF metrics endpoints, but there is no single Oshun admin web app and no dedicated Oshun admin mobile app for review queues, support, moderation, approvals, trust and safety, billing, entitlements, privacy requests, or on-call operations.
  9. The assistant/platform ownership model is still blurred. The repo indicates Iris should own persistent assistant data and memory, Psyche should own real-time multimodal/avatar orchestration, Lilith should own contemplative persona and meditation-tone logic, and the Oshun shell assistant should own product UX. Today those boundaries are not yet enforced cleanly.
  10. The repo has more safety, evaluation, watermarking, and abuse-prevention infrastructure than the earlier version of this report credited. Lilith in particular already includes substantive moderation, crisis, jailbreak, cloned-voice abuse, watermarking, and eval systems. The gap is less "no safety foundation" and more "no single cross-product release gate and review operating model."
  11. The local Civitai integration surface is broader than first assumed, but it should be treated carefully. The repo already includes Civitai providers, Studio routes, download/sync logic, and model-management docs. However, parts of that stack are still simulation-heavy or provider-assumptive, and current public documentation is clearest around model/library/discovery workflows, not around treating Civitai as Oshun's canonical generation runtime. Oshun should use Civitai primarily as a model-intelligence and asset-source layer, with Isis/ComfyUI/RunPod remaining the canonical execution and provenance plane.

Bottom-line recommendation#

OSHUN should be reframed and implemented as a contemplative intelligence product:

  • Tara = the experiential center
  • Arete = disciplined embodiment
  • Veritas = epistemic clarity
  • Nyx = awe and cosmological perspective
  • Nisaba = textual depth, lineage, and comparative wisdom
  • Sophia = grounding, citations, and research integrity behind the scenes

The next major product move should be:

  1. Expand the shell contract to recognize nisaba.
  2. Recenter the product around Tara-led rituals and enlightenment-oriented journeys.
  3. Introduce shared cross-domain objects and workflows so the domains stop feeling separate.
  4. Elevate Sophia into the canonical grounding and citation substrate for Veritas, Nisaba, assistant, and studio flows.
  5. Launch Nisaba in two layers:
    • web: serious research and reading workspace
    • mobile: study companion, saved passages, annotations, compare, concept threads, reminders
  6. Define Oshun Admin as a first-class companion product:
    • web: full operations, review, moderation, content ops, model ops, support, analytics, and governance
    • mobile: urgent review queues, approvals, escalation, incident response, on-call, creator/reviewer notifications
  7. Define a canonical assistant stack:
    • Iris = persistent memory and assistant state
    • Psyche = embodied real-time text/voice/avatar interaction
    • Lilith = contemplative teacher/persona logic and domain-specific safety
    • Sophia = grounding/citation substrate
    • Oshun shell = consumer UX, routing, and cross-domain orchestration

2. Research Method#

This report is grounded in two input classes:

2.1 Repository evidence#

Reviewed source and docs from:

  • docs/releases/v1/specs/product-brief.md
  • docs/releases/v1/scope/mvp-scope.md
  • docs/releases/v1/domains/nisaba/README.md
  • docs/releases/v1/domains/nisaba/feature-inventory.md
  • docs/releases/v1/domains/nisaba/implementation-map.md
  • docs/releases/v1/domains/iris/README.md
  • docs/domains/iris/features.md
  • docs/releases/v1/domains/sophia/README.md
  • docs/releases/v1/domains/sophia/implementation-map.md
  • docs/domains/sophia/architecture.md
  • docs/domains/sophia/database.md
  • docs/reference/api-contracts.md
  • docs/reference/data-ownership-matrix.md
  • docs/reference/monitoring.md
  • docs/releases/v1/specs/studio-domain-dependency-and-data-contract-map.md
  • docs/migration/yemaya-migration-guide.md
  • docs/domains/isis/README.md
  • docs/infrastructure/runpod-integration.md
  • docs/integrations/civitai/README.md
  • docs/integrations/civitai/api.md
  • docs/integrations/civitai/model-management.md
  • docs/integrations/civitai/training.md
  • docs/integrations/civitai/best-practices.md
  • docs/integrations/comfyui/README.md
  • docs/integrations/elevenlabs/README.md
  • docs/domains/lilith/extras/reports/PLATFORM_STATUS_REPORT.md
  • docs/domains/lilith/extras/reference/safety-measures.md
  • docs/domains/lilith/extras/reference/reporting-system.md
  • docs/domains/lilith/extras/reference/audio-api.md
  • apps/oshun/web/*
  • apps/oshun/mobile/*
  • apps/oshun/bff/*
  • libs/oshun/*
  • libs/nisaba/*
  • libs/contracts/iris/*
  • libs/contracts/psyche/*
  • apps/sophia/*
  • libs/sophia/*
  • libs/openapi/src/specs/sophia/sophia-api.yaml
  • libs/proto/src/sophia/sophia.proto
  • apps/yemaya/api/*
  • apps/lilith/*
  • apps/psyche/admin/*
  • apps/aphrodite/admin/*
  • apps/aphrodite/admin-bi-dashboard/*
  • services/psyche/*
  • libs/aphrodite/research-agents/*
  • apps/lilith/svc-ai/src/evals.ts
  • apps/lilith/svc-tts/src/audio-watermarking-system.ts
  • apps/lilith/svc-tts/src/abuse-risk-scoring-engine.ts
  • apps/lilith/svc-media/src/app.ts
  • apps/lilith/svc-ai/src/image-moderation/*
  • openapi/isis/openapi.yaml

2.2 External competitor and adjacent-product research#

Used current official product pages/docs/support pages on March 22, 2026 to identify strong ideas worth adapting:

  • Headspace
  • Calm
  • Insight Timer
  • Fabulous
  • Todoist
  • Duolingo
  • Google News
  • Ground News
  • Feedly
  • Stellarium
  • SkySafari / LiveSky
  • NASA Eyes / JPL Apps
  • Sefaria
  • Logos
  • Transkribus
  • Tropy
  • NotebookLM
  • Perplexity Deep Research
  • Elicit
  • Semantic Scholar
  • ComfyUI
  • RunPod
  • ElevenLabs

All external links are listed in the appendix.


3. Current-State Findings From the Repo#

3.1 What the Oshun shell currently exposes#

As of March 22, 2026, the Oshun web/mobile shell only exposes four domains:

  • tara
  • veritas
  • nyx
  • arete

This is hard-coded in the most important shared contracts:

  • apps/oshun/web/src/navigation/routes.ts
  • apps/oshun/mobile/src/navigation/routes.ts
  • libs/oshun/navigation/src/routes.ts
  • libs/oshun/domain-registry/src/index.ts
  • apps/oshun/bff/src/routes/feed-helpers.ts
  • libs/oshun/auth/src/types.ts
  • libs/oshun/shell-assistant/src/types.ts

This matters because Nisaba is not just absent from UI cards. It is absent from the shell's domain model itself.

3.2 Nisaba already exists in the repo, but outside the shell#

The repo already has substantial Nisaba implementation and documentation:

  • docs:
    • docs/releases/v1/domains/nisaba/README.md
    • docs/releases/v1/domains/nisaba/feature-inventory.md
    • docs/releases/v1/domains/nisaba/implementation-map.md
  • libraries:
    • libs/nisaba/annotations
    • libs/nisaba/assistant
    • libs/nisaba/canon
    • libs/nisaba/client
    • libs/nisaba/comparative
    • libs/nisaba/core
    • libs/nisaba/corpora
    • libs/nisaba/criticism
    • libs/nisaba/database
    • libs/nisaba/editions
    • libs/nisaba/geotemporal
    • libs/nisaba/languages
    • libs/nisaba/paleography
    • libs/nisaba/philology
    • libs/nisaba/schemas
    • libs/nisaba/standards
    • libs/nisaba/translations
    • libs/nisaba/workspace

The implementation-map doc explicitly says Nisaba has no Oshun app integration today. That aligns with the code search results.

3.3 Web is domain-deep; mobile is shell-first but domain-thin#

Web#

The web app has rich domain routing:

  • Tara subroutes for analytics, collections, courses, programs, search, sounds, teachers
  • Veritas subroutes for articles, bias, claims, fact-check, headlines, knowledge graph, newsletter, NLP, research
  • Nyx subroutes for analysis, catalogs, events, learning, moon, observation log, renderer, satellites, solar, sonification, star chart, telescope, time travel, widgets
  • Arete subroutes for affirmations, balance, coach, gamification, goals, habits, journal, seven habits, time, vision

This means the web app already behaves like multiple substantial domain products inside a shell.

Mobile#

The mobile app has:

  • a strong shell structure
  • unified tabs
  • domain launch route
  • shared discovery/home/activity/profile framing

But it does not have equivalent route-level depth per domain. Most domain depth is collapsed into app/domain/[domainId].tsx.

Implication:

  • web currently optimizes for depth
  • mobile currently optimizes for shell consistency
  • the product still lacks a unified cross-platform domain operating model

3.4 The current product brief still centers only four domains#

docs/releases/v1/specs/product-brief.md frames OSHUN as a unified product for exactly four domains. That brief is coherent for MVP, but it is now the main documentation constraint preventing Nisaba from being part of the product narrative.

This means the repo has a strategic mismatch:

  • product brief and shell contracts: four-domain worldview
  • actual repository capability inventory: at least one additional high-value domain already exists

3.5 Nisaba is not a small add-on#

The Nisaba docs and libraries show it is not a lightweight reading feature. It is a serious scholarly platform with:

  • ancient-language support
  • translation comparison
  • manuscript witness handling
  • collation and apparatus workflows
  • canon/reference systems
  • paleography
  • philology
  • comparative concept mapping
  • timeline and geotemporal tools
  • standards/export
  • workspace and research project tooling

This is why adding Nisaba as a simple fifth card would be strategically wrong. It needs a product role.

3.6 Sophia already exists and should be treated as substrate, not a naive sixth card#

The repo shows Sophia is already a substantial domain with a real architectural footprint:

  • apps/sophia/ingestion
  • apps/sophia/search-api
  • apps/sophia/knowledge-graph
  • apps/sophia/workbench
  • libs/sophia/database
  • libs/sophia/vectordb
  • libs/sophia/evaluation
  • libs/sophia/theory
  • libs/sophia/citation-analysis

Its docs explicitly position it as the knowledge backbone of the Oshun ecosystem. Its database schema models documents, chunks, citations, entity mentions, relations, knowledge packs, and ingestion jobs. Its current Oshun references are concentrated in Studio/Hathor research-grounding flows rather than the primary shell.

That creates an important product distinction:

  • Nisaba should become a user-facing depth domain in the main Oshun shell.
  • Sophia should primarily become the shared grounding, retrieval, citation, and knowledge-graph substrate for Nisaba, Veritas, assistant, and studio.
  • If Sophia gets direct UI prominence, it should be a web-first pro/workbench surface, not automatically a sixth consumer lifestyle card beside Tara, Arete, Veritas, Nyx, and Nisaba.

4. The Core Product Problem#

The current Oshun shell is unified structurally, but not yet unified philosophically.

4.1 What the shell currently does well#

  • Shared navigation exists.
  • Shared domain registry exists.
  • Shared auth, analytics, assistant, offline, and notification directions exist.
  • The home/explore/activity/profile model is coherent.
  • The product already thinks in cross-domain terms.

4.2 What the shell currently does not yet do well#

  • It does not provide a strong reason why all domains belong together beyond shared ownership.
  • It does not establish Tara as the governing rhythm of the product.
  • It does not yet turn discoveries in one domain into meaningful actions in another.
  • It does not yet expose a common underlying knowledge graph, ritual graph, or progress model.
  • It does not yet expose one clearly shared grounding and citation system for research-backed answers, claims, passages, and AI-generated summaries.
  • It does not yet give Nisaba a place in the user journey.

4.3 The consequence#

Without stronger cohesion, OSHUN risks becoming:

  • visually unified, but mentally fragmented
  • broad, but not meaningfully deep
  • cross-domain in navigation, but not cross-domain in value

The answer is not "merge everything into one feed." The answer is to give each domain a role inside one spiritual-intellectual operating system.


5. Proposed Product North Star#

5.1 Product framing#

OSHUN should be framed as:

A contemplative intelligence system for living with more clarity, depth, discipline, wonder, and wisdom.

That framing makes the domains coherent:

  • Tara helps the user become still and present.
  • Arete helps the user embody practice in daily life.
  • Veritas helps the user see clearly and resist confusion.
  • Nyx helps the user recover awe, scale, and cosmic humility.
  • Nisaba helps the user trace wisdom traditions, sources, lineages, and meanings deeply.

5.2 Tara should be the primary theme#

If meditation and enlightenment are the main theme, Tara should not just be "the meditation section." Tara should be the OSHUN-wide experiential layer for:

  • beginning the day
  • pausing before reactive behavior
  • processing difficult information
  • integrating learning
  • transitioning into sleep
  • framing personal growth as consciousness work, not mere productivity

That gives the product a center of gravity.

5.3 Domain role model#

Recommended permanent role model:

Domain Product role Primary emotional function Primary functional mode
Tara The center Calm, presence, awakening Ritual, session, reset, reflection
Arete The embodiment layer Discipline, agency, integrity Goal/habit/journal/coach execution
Veritas The clarity layer Trust, discernment, reality contact Explain, verify, compare, track
Nyx The awe layer Wonder, scale, humility Observe, time, learn, experience
Nisaba The depth layer Meaning, lineage, scholarship Read, compare, annotate, trace

This structure makes Nisaba feel native instead of bolted on.

Additional platform substrate:

  • Sophia = the grounding layer. Emotional function: trustworthiness. Functional mode: ingest, search, cite, verify, graph, evaluate. Sophia should usually surface through product capabilities rather than as a default consumer card.

6. Cohesion Gap Analysis#

6.1 Gap A: Narrative cohesion gap#

Current state#

The shell says "one product," but the domains still read as adjacent properties.

Why it matters#

Users need a reason to believe Tara, Arete, Veritas, Nyx, and Nisaba belong together. Without a unifying philosophy, cross-domain suggestions feel arbitrary.

Recommendation#

Make the main product narrative:

  • wisdom in practice
  • truth in context
  • awe in daily life
  • calm in the face of complexity

6.2 Gap B: IA and workflow cohesion gap#

Current state#

The shell has top-level tabs, but each domain still behaves like a separate island after launch.

Missing#

  • shared ritual engine
  • shared collections/workspaces
  • shared progress graph
  • shared concept graph
  • shared "continue" semantics beyond simple recency

Recommendation#

Introduce global objects that can connect all domains:

  • Ritual
  • Session
  • Insight
  • Claim
  • Source
  • Observation
  • Passage
  • Concept
  • Project
  • Collection

This is the real cohesion layer.

6.3 Gap C: Context transfer gap#

Current state#

Cross-domain context mostly transfers through navigation, not meaning.

Desired state#

Examples:

  • a Tara session on impermanence can lead to Nisaba passages on Buddhist, Stoic, Taoist, and Christian parallels
  • a Veritas claim about a historical text can open Nisaba source lineage and manuscript context
  • a Nyx celestial event can trigger a Tara night-sky meditation and an Arete reflection prompt
  • an Arete burnout pattern can suggest Tara breathing, Veritas input reduction, and Nyx evening awe exposure

6.4 Gap D: Web/mobile role clarity gap#

Current state#

  • web has more domain depth
  • mobile has more shell polish

Problem#

The product is not explicitly deciding which surfaces are:

  • mobile-first
  • web-first
  • parity-required
  • companion-only

Recommendation#

Define per-domain platform intent:

  • Tara: parity-required
  • Arete: parity-required
  • Veritas: parity-required for core flows, research depth stronger on web
  • Nyx: parity-required for event and observation loops, richer visualization on web/PWA and sensor-native behaviors on mobile
  • Nisaba: mobile companion + web powerhouse

6.5 Gap E: Definition-of-complete gap#

The repo has many domain assets, but no cross-domain definition of what "complete and integrated" means.

Recommended definition of done for any Oshun domain:

  1. Discoverable in home/explore/search/assistant.
  2. Launchable through the shared route and entitlement model.
  3. Has save/library semantics.
  4. Has activity timeline semantics.
  5. Has notification and reminder semantics where relevant.
  6. Has cross-domain recommendation hooks.
  7. Has shared analytics coverage.
  8. Has graceful offline/degraded behavior.
  9. Has mobile/web role clarity.
  10. Has at least three meaningful bridges into other domains.
  11. If it makes factual claims or relies on AI generation, it has provenance, citation, and evaluation semantics.

By that definition, Nisaba is currently incomplete. So are parts of the other four.


7. Domain Readiness Assessment#

Suggested maturity ladder:

  • L0: invisible to shell
  • L1: launchable
  • L2: useful in isolation
  • L3: cross-domain aware
  • L4: role-clear, habit-forming, and deeply integrated
Domain Current rating Why
Tara L3 Strong launch presence and meaningful surfaces, but not yet the OS-level ritual spine
Veritas L3 Rich features and shell value, but still too isolated from Tara/Nisaba meaning context
Nyx L2-L3 Strong experiential depth on web and promising logic, but less integrated into daily product habits
Arete L2-L3 Rich underlying logic and many shell surfaces, but still not fully grounded in shared workflows and persistence reality
Nisaba L0 Powerful libraries/docs exist, but no shell exposure in web/mobile
Sophia L2-L3 platform / L0 consumer Strong research backbone with apps, database, vector, citation, and evaluation systems, but not yet productized as Oshun's canonical grounding substrate or a coherent end-user surface

Target state:

  • Tara: L4
  • Arete: L4
  • Veritas: L4
  • Nyx: L4
  • Nisaba: L3 initially, then L4 on web and L3-L4 on mobile companion flows
  • Sophia: L4 as a platform substrate, L1-L2 only as a direct pro/workbench surface

8. Domain-by-Domain Analysis#

8.1 Tara#

Current repo state#

Strengths:

  • strong product identity
  • meaningful web route depth
  • mobile shell integration
  • reusable audio/session/favorites/course primitives in @oshun/domain-tara
  • existing audit confirms valuable reusable patterns from the standalone Tara app

Gaps:

  • Tara is still mostly "the meditation domain," not the product-wide ritual layer
  • Tara is not yet clearly orchestrating start-of-day, mid-day reset, news decompression, reflection, or sleep transitions across the shell
  • Tara's cross-domain value is underspecified

Relevant competitor ideas#

  • Headspace: progressive guided journeys, mind/body integration, and structured daily practice flows
    Source: https://help.headspace.com/hc/en-us/articles/360043063714-What-is-Move-Mode
  • Calm: flagship daily editorial ritual, sleep stories, music, and mindful movement across devices
    Source: https://support.calm.com/hc/en-us/articles/360008536834-Calm-Premium-vs-Free-Features-Content-List-Benefits
  • Insight Timer: live events, teacher/community network, and a sense of a living practice ecosystem
    Source: https://help.insighttimer.com/support/solutions/articles/67000665284-applying-for-live-events-creating-a-live-event-proposal

Best ideas to adapt#

  • A single daily anchor experience.
  • Morning, mid-day, and evening session slots.
  • Pre-sleep and post-news decompression flows.
  • Live or scheduled shared practice moments.
  • Teacher-led paths, but inside Oshun's unified identity.

Tara's required role in a unified OSHUN#

Tara should become the shell's temporal orchestrator:

  • morning opening ritual
  • stress interruption
  • transition between deep cognitive modes
  • reflection close
  • sleep descent

Tara-specific integration requirements#

  • feed recommendations into Arete routines
  • consume Veritas signals to trigger "pause before reacting" flows
  • pair Nyx events with awe meditation or observation rituals
  • use Nisaba passages as contemplative text companions

Tara completion definition#

Tara is complete only when:

  • it is the default reset mechanism across OSHUN
  • every other domain has at least one Tara bridge
  • the home dashboard feels Tara-guided even when the user is not explicitly "inside Tara"

8.2 Arete#

Current repo state#

Strengths:

  • very deep domain libraries under libs/arete/*
  • strong conceptual coverage: goals, habits, journal, vision, time, balance, affirmations, AI coach, gamification, seven habits
  • existing Oshun surfaces already use parts of Arete

Gaps:

  • no standalone Arete app shell to inherit from
  • much of Arete still needs persistence/repository integration
  • some logic is placeholder/simulated
  • Arete can easily drift into generic productivity without Oshun's contemplative frame

Relevant competitor ideas#

  • Fabulous: behavior-science journeys and routine-building narratives
    Source: https://www.thefabulous.co/
  • Todoist: recurring tasks, daily goals, karma, streaks, strong completion feedback, and lightweight accountability
    Source: https://www.todoist.com/de/karma
  • Duolingo: mastery loops, streak psychology, challenge cadence, and playful reinforcement
    Source: https://blog.duolingo.com/de/streak-feiern/

Best ideas to adapt#

  • ritualized habit onboarding instead of generic task creation
  • explicit streak/goal feedback
  • daily quest framing
  • visual progress/status loops
  • immediate next action clarity

Required product correction#

Arete must not become a detached productivity tool. It should be:

  • contemplative discipline
  • ethical embodiment
  • practice continuity

In other words: Arete should be Tara enacted.

Arete-specific integration requirements#

  • Tara should define state-of-being inputs for Arete plans
  • Veritas should feed research/reading goals into Arete focus blocks
  • Nyx should create seasonal or event-based reflection rituals
  • Nisaba should power study plans, source journaling, and lineage-based practice prompts

Arete completion definition#

Arete is complete when:

  • goals/habits/journal exist as one loop, not separate tools
  • it is visibly tied to Tara's contemplative intent
  • cross-domain actions can become structured commitments in one tap

8.3 Veritas#

Current repo state#

Strengths:

  • broad feature surface on web
  • strong shell relevance
  • strong reusable infra from standalone mobile audit
  • clear role in shared search, saved items, and notifications

Gaps:

  • not enough integration with Tara as a psychological regulator
  • not enough connection to Nisaba as a source/provenance/primary-text layer
  • can feel like a news product living beside a mindfulness product, instead of inside a truth-and-wisdom system

Relevant competitor ideas#

  • Google News Full Coverage: story clustering, evolving-topic context, explainers, and local angles
    Source: https://blog.google/products-and-platforms/products/news/get-full-news-story-full-coverage-search/
  • Ground News: bias distribution, blindspot framing, cross-source comparison, and explicit lens awareness
    Source: https://ground.news/media-bias
  • Feedly: AI feeds, source-following, boards, saved research structures, and organized workspaces
    Source: https://docs.feedly.com/article/801-how-to-navigate-through-feedly

Best ideas to adapt#

  • story arcs instead of flat headlines
  • explainability and lens diversity
  • user-controlled research boards
  • strong saved-source and topic tracking
  • signal-before-noise ranking

Required product correction#

Veritas should not be just "news inside Oshun." It should be Oshun's epistemic hygiene system.

That means Veritas needs bridges to:

  • Tara: calm before reaction
  • Arete: action based on informed commitment
  • Nisaba: primary source depth and textual lineage
  • Nyx: perspective, scale, scientific wonder

Veritas completion definition#

Veritas is complete when:

  • it not only informs but helps the user metabolize information wisely
  • major claims can route to primary texts and deeper source context where applicable
  • the product can move a user from confusion to grounded clarity

8.4 Nyx#

Current repo state#

Strengths:

  • large and impressive astronomy surface on web
  • meaningful simulation and render logic
  • promising offline/sync/widget direction
  • strong emotional distinctiveness

Gaps:

  • still not strongly woven into daily practice
  • some placeholder/simulated data paths remain
  • its role in the shell can feel like a cool side domain instead of a core part of the product's worldview

Relevant competitor ideas#

  • Stellarium: realistic sky representation and rich sky-culture framing
    Source: https://stellarium.org/en/landscapes-south-america.html
  • SkySafari / LiveSky: observing lists, cloud sync, logs, equipment, sites, sessions, and community analytics
    Source: https://www.skysafariastronomy.com/livesky-data-management-and-cloud-sync-service.html
  • NASA Eyes / JPL Apps: authoritative interactive 3D solar-system exploration and time-based cosmic navigation
    Source: https://www.jpl.nasa.gov/apps

Best ideas to adapt#

  • observation lists and logging
  • event calendars with meaningful reminders
  • "time travel" and orbital context
  • shared saved objects and sessions
  • cultural/civilizational sky narratives

Required product correction#

Nyx should be positioned as:

  • the domain of awe
  • the cosmic reminder against ego and panic
  • a natural companion to Tara's stillness

Nyx is one of the strongest ways to make "enlightenment" concrete and embodied in the product without sounding abstract.

Nyx-specific integration requirements#

  • Tara: night-sky meditation, moon-phase reflection, awe-based calming rituals
  • Arete: observation habits, seasonal practices, outdoor routines, sleep hygiene support
  • Veritas: scientific explainers, evidence-backed astronomy learning
  • Nisaba: ancient cosmologies, star texts, calendars, sky symbolism, comparative interpretations

Nyx completion definition#

Nyx is complete when:

  • it participates in daily ritual and seasonal rhythm
  • it offers both wonder and practice
  • it functions as a bridge between embodied life and cosmic perspective

8.5 Nisaba#

Current repo state#

This is the clearest gap and biggest strategic opportunity.

Strengths already in-repo:

  • huge scope already modeled in docs
  • serious library footprint under libs/nisaba/*
  • research-environment support for:
    • project management
    • split-pane reading
    • cross-corpora search
  • client/hooks support for:
    • manuscripts
    • collation
    • annotations
    • editions
    • references
    • corpus search
    • concepts
    • translations
    • morphology
    • lexicon
    • timeline
    • influence network
    • realtime

Gaps:

  • no shell exposure in web/mobile
  • no shared @oshun/domain-nisaba adapter package
  • no route inclusion
  • no entitlement, analytics, notification, search, assistant, or BFF inclusion
  • no clear consumer-facing role definition inside OSHUN

Relevant competitor ideas#

  • Sefaria: interconnected texts, related-text side panels, adjacent reading, and a living library model
    Source: https://help.sefaria.org/hc/en-us/articles/18613227644316-How-to-Find-Interconnected-Texts
  • Logos: linked original-language and lexicon workflows, factbook-style study, and pane coordination
    Source: https://support.logos.com/hc/en-us/articles/28716717196813-Study-a-Lemma-with-Logos
  • Transkribus: handwriting recognition, custom models, document search, TEI/XML export, and collaborative collections
    Source: https://www.transkribus.org/handwriting-to-text
  • Tropy: disciplined research-image metadata, tagging, and archival organization
    Source: https://docs.tropy.org/before-you-begin/metadata

Strategic role inside OSHUN#

Nisaba should not be positioned as a niche academic appendix. It should be positioned as:

  • the deep-source layer
  • the lineage layer
  • the text and meaning layer
  • the place where ancient wisdom, interpretation, scholarship, and evidence meet

Consumer-facing vs expert-facing split#

Nisaba must have two modes:

Consumer-facing mode#

  • daily passage
  • compare translations
  • see parallel concepts across traditions
  • save, highlight, annotate
  • connect passages to Tara reflections and Arete journaling

Expert-facing mode#

  • manuscript witness work
  • collation
  • lexicon/morphology
  • scholarly workspace
  • cross-tradition concept graph
  • standards-aware export

Nisaba-specific integration requirements#

  • Tara:
    • contemplative readings
    • koan/sutra/verse companions
    • guided reflection after passages
  • Arete:
    • study habit loops
    • journaling prompts from texts
    • virtue or discipline tracks grounded in primary sources
  • Veritas:
    • provenance, textual lineage, and primary-source depth
    • bridge from claims to source texts
  • Nyx:
    • ancient astronomy texts
    • calendars, cosmologies, symbolic sky traditions

Nisaba completion definition#

Nisaba is complete when:

  1. it is visible in the shell;
  2. it has web and mobile roles that are different but coherent;
  3. it is connected to search/library/activity/assistant/notifications;
  4. it adds meaningful depth to the other four domains instead of competing with them.

8.6 Sophia#

Current repo state#

Strengths:

  • Sophia is explicitly documented as the knowledge backbone of the Oshun ecosystem.
  • It has real application boundaries:
    • ingestion
    • search API
    • knowledge graph
    • web workbench
  • It has a real data model under libs/sophia/database, including documents, chunks, citations, entity mentions, relations, knowledge packs, and ingestion-job tracking.
  • It has retrieval, vector, citation-analysis, theory, and evaluation libraries that are directly relevant to grounded answers and serious research tooling.
  • Oshun already references Sophia in Studio/Hathor research-grounding routes, which means some product adjacency is already established.

Gaps:

  • Sophia is not yet clearly classified as either:
    • a top-level end-user domain
    • or the mandatory shared substrate for grounded research across Oshun
  • Parts of its runtime remain memory-first:
    • the knowledge-graph app uses a CombinedMemoryStore
    • citation-manager defaults include InMemoryCitationStorage
  • Veritas, Nisaba, assistant, and other shells are not visibly forced through a common Sophia-backed grounding contract.
  • There is no product-wide requirement that factual claims, generated research reports, or comparative-text explanations expose citation trails via Sophia.

Relevant competitor ideas#

  • NotebookLM: source-grounded notebooks, auto-summaries, and question answering based on supplied sources Source: https://blog.google/innovation-and-ai/products/notebooklm-new-features-availability/
  • Perplexity Deep Research: iterative multi-source research, report writing, and export/share flows Source: https://www.perplexity.ai/es-es/hub/blog/introducing-perplexity-deep-research
  • Elicit: structured literature review, screening, data extraction, research reports, and transparent intermediate steps Source: https://support.elicit.com/en/articles/7927169
  • Semantic Scholar: literature graph, APIs/datasets, and machine-readable scholarly infrastructure Source: https://www.semanticscholar.org/faq/open-data

Best ideas to adapt#

  • source-grounded notebook/workspace behaviors
  • transparent research steps and intermediate artifacts
  • exportable evidence tables and bibliography outputs
  • claim-to-source inspectability
  • graph-first discovery across authors, concepts, traditions, and citations

Required product correction#

Sophia should usually not be treated as "another lifestyle tab."

The better model is:

  • Sophia is the product-wide grounding substrate.
  • Nisaba is the user-facing deep text/lineage/study domain.
  • Veritas is the user-facing clarity/news/claims domain.
  • Assistant and Studio consume Sophia to provide citation-backed answers, notebooks, concept graphs, and research reports.

If Sophia gets a branded first-class surface, it should be:

  • web-first
  • pro/researcher/creator-oriented
  • more like a workbench than a mass-consumer home tab

Sophia-specific integration requirements#

  • Tara:
    • teacher/source provenance
    • citation-backed contemplative readings
    • grounded explainers for practices when appropriate
  • Arete:
    • evidence-backed study plans
    • reference-linked journaling and habit prompts
    • notebooks for longer transformation arcs
  • Veritas:
    • canonical claim grounding
    • inspectable citation trails
    • source quality scoring and concept clustering
  • Nyx:
    • scientific explainers
    • historical sky-text research
    • source packs for cosmology and calendars
  • Nisaba:
    • cross-text search
    • bibliography and citation management
    • concept graph and notebook substrate

Sophia completion definition#

Sophia is complete when:

  1. it is the canonical grounding API for the Oshun ecosystem;
  2. assistant answers can produce inspectable citation trails by default;
  3. Veritas and Nisaba share a common source, citation, and concept graph model;
  4. workbench/studio research flows can export durable evidence artifacts; and
  5. any direct Sophia UI is clearly positioned as a web-first pro surface rather than a confusing extra consumer tab.

9. Web vs Mobile Gap Analysis#

Area Web current state Mobile current state Gap Recommendation
Domain depth high medium-low parity mismatch decide domain-by-domain platform role, not blanket parity
Home/dashboard rich shell and domain modules strong shell cards and quick actions still card-centric evolve into ritual-centric orchestration
Domain routing many per-domain subroutes single generic domain route mobile depth bottleneck create domain-specific feature modules over time
Search/discovery cross-domain but largely shell utility cross-domain but lighter meaning/context not deep enough add concept, ritual, passage, source, and project entities
Notifications unified structure unified structure mostly domain-by-domain alerts add ritual reminders and synthesis digests
Saved/library shared direction exists shared direction exists library model still shallow unify around collection, passage, article, observation, routine, goal, source
Assistant strong potential less explicit product centrality underused as cohesion layer make assistant the cross-domain meaning broker
Nisaba none none complete absence launch web-first + mobile-companion Nisaba MVP
Sophia studio/research workbench only none or highly indirect user-facing role unresolved keep Sophia web-first/pro-facing; expose grounded answers, notebooks, and citations across both apps rather than adding a raw shell card
Cross-domain context limited limited mostly nav-based build shared context graph and journey chaining

Practical interpretation#

The correct goal is not "make mobile look like web." The correct goal is:

  • one product logic
  • one domain model
  • one shared memory and context model
  • platform-appropriate surface depth

10. What a Truly Unified OSHUN Should Feel Like#

10.1 Daily flow model#

Recommended default daily flow:

Morning#

  • Tara: short centering
  • Arete: daily commitments
  • Veritas: one calm, high-signal briefing
  • Nyx: sky/weather/cosmic note when relevant
  • Nisaba: one passage or concept thread

Midday#

  • Tara: reset
  • Arete: focus block check-in
  • Veritas: update only if high significance
  • Nisaba: optional study continuation

Evening#

  • Nyx: tonight's sky / moon / celestial event
  • Tara: decompress or sleep flow
  • Arete: journal and reflection
  • Nisaba: text reflection or annotation

This makes the domains feel like phases of one life practice.

10.2 Shared cross-domain objects#

Recommended global object model:

  • Ritual: one or more steps across domains
  • Collection: saved mix of articles, passages, meditations, objects, goals
  • Concept: shared semantic node across Veritas and Nisaba, optionally Nyx and Tara
  • Practice: repeatable behavior sequence from Tara + Arete
  • Insight: distilled cross-domain result that can be saved/shared/journaled
  • Evidence: source-backed support object for claims, passages, and reports
  • Citation: reusable provenance link between claims, sources, and generated outputs
  • Notebook: mixed research and reflection workspace across Nisaba, Veritas, Sophia, and Studio
  • ResearchReport: structured synthesis artifact with sources and confidence
  • Project: deeper study or improvement arc

10.3 Cross-domain examples that should exist#

  1. A user saves a Veritas article on attention economy harms. The product suggests:

    • Tara breathing/reset
    • Nisaba passages on attention, desire, silence, or right speech
    • Arete focus routine changes
  2. A user opens a Nyx meteor event. The product suggests:

    • a Tara awe meditation
    • an Arete reflection prompt
    • a Nisaba card on ancient interpretations of the sky
  3. A user studies a Nisaba passage on compassion. The product suggests:

    • Tara compassion session
    • Arete journaling prompt
    • Veritas reading on contemporary altruism or ethics

These are the kinds of journeys that make OSHUN feel singular.


11. Competitor Synthesis by Domain#

11.1 Tara synthesis#

What the best competitors do well:

  • consistent daily ritual slot
  • sleep and rest positioning
  • movement and breathing as companions, not separate products
  • community and teacher energy

What OSHUN should do differently:

  • bind meditation to truth, scholarship, and awe
  • move beyond wellness into contemplative wisdom
  • use Tara as the organizing center for the entire product

11.2 Arete synthesis#

What the best competitors do well:

  • habit onboarding
  • clear completion feedback
  • streak loops
  • challenge framing

What OSHUN should do differently:

  • avoid generic productivity culture
  • tie every habit/goal to meaning and presence
  • connect action plans to wisdom and reflection

11.3 Veritas synthesis#

What the best competitors do well:

  • story clustering
  • viewpoint comparison
  • AI/source organization
  • research workspace behaviors

What OSHUN should do differently:

  • pair epistemic clarity with emotional regulation and deep source context
  • reduce reactive consumption
  • surface primary-source depth where possible

11.4 Nyx synthesis#

What the best competitors do well:

  • realism
  • observation tooling
  • logs and sync
  • 3D exploration

What OSHUN should do differently:

  • make astronomy part of an awakening-oriented life practice
  • connect sky events to contemplation and seasonal ritual
  • bring ancient cosmology and textual interpretation in through Nisaba

11.5 Nisaba synthesis#

What the best competitors do well:

  • linked reading and interconnections
  • original language depth
  • manuscript and handwriting tooling
  • metadata discipline

What OSHUN should do differently:

  • make scholarly depth available without making the main product academic-only
  • bridge primary texts directly into meditation, journaling, verification, and awe
  • support both public meaning-making and expert scholarship

11.6 Sophia synthesis#

What the best competitors do well:

  • source-grounded notebook behavior
  • autonomous but inspectable research workflows
  • transparent intermediate steps
  • literature-graph and citation-network navigation

What OSHUN should do differently:

  • bind research grounding to contemplative product contexts, not just academic workflows
  • make the same citation substrate power Veritas, Nisaba, assistant, and studio
  • avoid a black-box "AI answer" experience by making provenance first-class

12. Nisaba Integration Strategy#

12.1 The wrong way to add Nisaba#

Do not:

  • add a fifth icon and stop
  • expose Nisaba without search/library/activity integration
  • make Nisaba purely academic and detached from the rest of the product
  • try to mirror full scholarly depth on mobile immediately

12.2 The right way to add Nisaba#

Add Nisaba in three layers.

Layer 1: Shell recognition#

Required code/system areas to expand:

  • shared navigation domain enum
  • web domain list and route validation
  • mobile domain list and route validation
  • domain registry metadata
  • auth and entitlement domain types
  • analytics domain enums
  • assistant domain maps
  • BFF feed/search/notification domain types
  • design tokens and domain iconography
  • domain switchers and filters

This is the minimum for shell legitimacy.

Layer 2: Launchable app surfaces#

Web Nisaba MVP#

  • Nisaba landing/dashboard
  • split-pane reader
  • compare translations
  • related concepts / parallel texts
  • saved passages and annotations
  • research workspace entry

Mobile Nisaba MVP#

  • daily passage
  • saved texts
  • quick compare
  • notes/highlights
  • study reminders
  • concept thread cards
  • "continue reading" and "continue study" shell hooks

Layer 3: Cross-domain role activation#

  • Tara receives contemplative passage companions.
  • Arete receives study habits and journal prompts.
  • Veritas receives primary-source context.
  • Nyx receives cosmology/calendar/textual overlays.

Only after layer 3 does Nisaba become truly integrated.


13. Proposed Nisaba Feature Scope by Platform#

13.1 Web-first features#

These should be Nisaba's strongest early surfaces:

  • split-pane reading
  • translation comparison
  • lexicon/morphology panes
  • concept network
  • manuscript/edition/project workspace
  • standards/export
  • research collections and projects

Rationale:

  • larger screen
  • better suited to serious study
  • aligns with existing libs/nisaba/workspace architecture

13.2 Mobile-first or mobile-companion features#

  • daily text
  • passage save/highlight
  • light annotation
  • compare a few translations
  • quick source lookup
  • concept cards
  • study reminders
  • "continue project" companion actions

Rationale:

  • supports daily return behavior
  • complements Tara and Arete loops
  • avoids forcing advanced scholarly UI into a cramped surface too early

13.3 Parity-required shared capabilities#

  • save
  • library
  • search
  • notifications
  • activity
  • assistant awareness
  • deep links
  • entitlements
  • analytics

14.1 Ritual engine#

Build a shared ritual system that can compose steps from multiple domains.

Example ritual types:

  • Morning Clarity
  • Evening Reflection
  • Truth Without Reactivity
  • Moon and Mind
  • Passage to Practice

Each ritual can include:

  • Tara session
  • Arete commitment/journal
  • Veritas briefing
  • Nyx observation
  • Nisaba reading

14.2 Unified library and collections#

Create one shared library model with typed items:

  • meditation
  • article
  • source
  • sky object
  • observation
  • goal
  • journal entry
  • passage
  • annotation
  • concept
  • project

This is one of the most important missing cohesion layers.

14.3 Shared concept graph#

This is especially important after Nisaba arrives.

Example concept nodes:

  • compassion
  • impermanence
  • truth
  • attention
  • humility
  • suffering
  • discipline
  • wonder

Then:

  • Tara can attach sessions to concepts
  • Arete can attach practices to concepts
  • Veritas can attach claims/topics to concepts
  • Nyx can attach events/themes to concepts
  • Nisaba can attach passages and traditions to concepts

14.4 Assistant as the cross-domain interpreter#

The assistant should become the product's "meaning layer."

Examples:

  • "Give me a calm but informed briefing."
  • "Show me texts from different traditions on attachment."
  • "Turn this article into a reflection practice."
  • "Pair tonight's moon with a short meditation and a journal prompt."

This is where OSHUN can become uniquely cohesive.

14.5 Evidence and provenance backbone#

Build one shared grounding system, powered by Sophia, for:

  • citations
  • source packs
  • claim inspection
  • notebook/research report export
  • confidence and groundedness signals

This should be visible anywhere Oshun explains, compares, summarizes, or teaches.

14.6 Generation control plane#

Build one shared generation control plane, led by Isis, for:

  • workflow definitions
  • workflow versioning
  • model/provider selection policy
  • lineage and provenance
  • output retention
  • quality-evaluation hooks

Tara audio, Psyche voice/avatar systems, Yemaya authoring flows, and any Oshun studio generation surface should consume this plane instead of each inventing its own provider contract.

14.7 Unified admin and operator surface#

Oshun should have dedicated admin products, not only scattered domain-specific workbenches.

Recommended split:

  • Oshun Admin Web:
    • trust and safety review
    • content/editorial review
    • persona and avatar management
    • model registry and provider operations
    • rights/license/provenance review
    • user support, entitlements, billing, and privacy operations
    • incident command, analytics, and audit/compliance views
  • Oshun Admin Mobile:
    • urgent queue triage
    • approval/reject/escalate actions
    • on-call alerts and incident acknowledgment
    • creator/reviewer notifications
    • limited but high-trust moderation and support actions

These admin surfaces should also expose internal AI copilots:

  • reviewer copilots
  • support copilots
  • research verification copilots
  • moderation explainers
  • model-ops assistants

But those assistants should remain auditable, disclosure-forward, and human-supervised.


15. Technical Integration Gaps for Nisaba#

The repo shows that adding Nisaba safely is a shell-wide contract change, not a small app-local patch.

15.1 Shared contract gaps#

Current four-domain assumptions exist in:

  • libs/oshun/navigation
  • libs/oshun/domain-registry
  • libs/oshun/auth
  • libs/oshun/shell-assistant
  • libs/oshun/shell-achievements
  • libs/oshun/design-tokens
  • libs/oshun/shell-desktop
  • libs/oshun/shell-wearable

15.2 App-layer gaps#

Web:

  • apps/oshun/web/src/navigation/routes.ts
  • apps/oshun/web/src/app/domains/[domainId]/page.tsx
  • apps/oshun/web/src/components/DomainRouteExperience.tsx
  • many domain label/accent/filter arrays across search, notifications, cross-domain panels, cards, routines, and onboarding

Mobile:

  • apps/oshun/mobile/src/navigation/routes.ts
  • apps/oshun/mobile/app/domain/[domainId].tsx
  • apps/oshun/mobile/src/components/DomainSwitcherEntry.tsx
  • analytics and notification types
  • search/recommendation/filter domain unions

15.3 BFF gaps#

  • domain IDs in feed helpers
  • likely search/filter/notification route contract expectations
  • any authorization and home/continue APIs returning domain arrays

15.4 Missing adapter package#

There is no libs/oshun/domain-nisaba.

Recommended creation:

  • libs/oshun/domain-nisaba

Responsibilities:

  • consumer-facing Nisaba card models
  • launch actions
  • deep links
  • lightweight mobile-friendly summary types
  • hooks/adapters over @nisaba/client and @nisaba/workspace

15.5 Missing product contract updates#

Needed additions:

  • Nisaba domain metadata
  • icon, accent, taglines, CTA, permissions, offline behavior
  • entitlement tier decision
  • search result item typing
  • notification labeling
  • assistant intent mapping
  • activity timeline typing
  • collections/library typing

16. Backing Systems Audit: Sophia, Assistants, Admin, and Media Generation#

The user asked for a comprehensive audit of the systems underneath the domains, not just the shell surfaces. That is the right scope. Oshun cannot become a cohesive contemplative intelligence product if its research, generation, authoring, and assistant layers remain fragmented or prototype-shaped.

16.1 System map and intended ownership#

The repo currently implies this capability map:

  • Iris:
    • canonical assistant identity, conversations, memory, and privacy boundary
  • Sophia:
    • canonical research/grounding/citation/knowledge-graph layer
  • Isis:
    • canonical generation factory and workflow control plane
  • Yemaya:
    • creative authoring, rendering, composition, and creation UX
  • Lilith:
    • meditation/audio/voice service implementations and generation adapters
  • Psyche:
    • live multimodal assistant, avatar, voice, session orchestration
  • Aphrodite:
    • experimental/autonomous research and automation agents

That overall shape is promising. The main problem is not lack of breadth. The main problem is inconsistent production maturity and blurred boundaries.

16.2 Sophia assessment: strongest candidate for the grounding substrate#

Sophia is comparatively well-formed as infrastructure.

Strengths:

  • strong architectural clarity in docs
  • explicit contract-first posture in docs/reference/api-contracts.md, including OpenAPI and protobuf coverage for Sophia
  • explicit apps for ingestion, search API, knowledge graph, and workbench
  • concrete PostgreSQL/Prisma schema for documents, chunks, citations, entities, relations, and ingestion jobs
  • vector-store abstraction with multiple backends
  • explicit retrieval evaluation, citation analysis, and theory frameworks
  • clear fit as the platform that should make Oshun answers more trustworthy

Gaps:

  • some runtime implementations are still memory-first
  • Oshun product surfaces do not yet consistently consume Sophia as a mandatory grounding layer
  • no single shared API contract makes "grounded answer with citations" a product-wide default behavior
  • Sophia Workbench review flows still include development/mock fallback paths, which weakens its claim as the already-hardened review backbone

Recommendation:

  • Sophia should become the mandatory substrate for any Oshun experience that:
    • makes claims
    • explains sources
    • synthesizes research
    • generates study notebooks or evidence reports
  • Sophia's OpenAPI/proto contracts should become the canonical contract that BFFs, shells, and admin tools consume for evidence, citations, source packs, and grounded reports

16.3 Assistant, memory, and embodied persona audit#

This is one of the biggest correctness gaps in the prior version of the report. The repo does not point to one assistant system. It points to at least four distinct layers that need to be separated intentionally.

Oshun shell assistant#

What exists:

  • libs/oshun/shell-assistant/* provides the shell-local assistant engine, routing model, and domain adapters.
  • apps/oshun/bff/src/routes/assistant.ts exposes the shell-facing API.
  • apps/oshun/web/src/components/assistant/AssistantPanel.tsx provides the current global assistant UI.

Current gap:

  • The default session store is in-memory.
  • The current web voice experience relies on browser SpeechRecognition / webkitSpeechRecognition and speechSynthesis, which is acceptable for prototyping but not for a flagship, cross-device embodied assistant.
  • Domain coverage still assumes only the four currently exposed Oshun domains.

Iris#

What exists:

  • docs/reference/data-ownership-matrix.md explicitly assigns AI assistant, conversations, and memory to Iris.
  • docs/releases/v1/domains/iris/README.md and docs/domains/iris/features.md position Iris as a universal assistant platform with hierarchical memory, multimodal interaction, privacy controls, and cross-device continuity.
  • libs/contracts/iris/* gives Iris a real contract surface for conversation, memory, and agent semantics.

Implication:

  • Iris should own persistent memory, user memory controls, memory consent, and assistant continuity across web/mobile/admin surfaces.

Psyche#

What exists:

  • services/psyche/README.md and the orchestrator docs describe a serious live-session platform spanning avatar, behavior, voice, persona, and orchestration.
  • Tavus-related services and libraries indicate strong investment in animated avatar, conferencing, turn-taking, and persona infrastructure.
  • The repo also includes Psyche Admin, which is already more advanced than Oshun's current consumer assistant operations tooling.

Implication:

  • Psyche should own real-time multimodal session orchestration, avatar runtime, conferencing bridges, and low-latency voice/avatar embodiment.

One repo inconsistency to resolve explicitly:

  • docs/reference/data-ownership-matrix.md describes Psyche differently from the Psyche domain docs/services. The implementation and domain-reference evidence lean strongly toward Psyche as the live virtual-assistant/embodiment layer, while assistant memory ownership still points more cleanly to Iris. The platform ownership map needs to make that distinction explicit.

Lilith#

What exists:

  • Lilith includes persona APIs, chat APIs, meditation experience logic, and stylized avatar/metaverse systems.
  • Lilith also includes explicit teacher/personality-registry work and a richer mobile persona/chat product surface than the current unified Oshun mobile shell.
  • Lilith also has stronger-than-credited safety/eval/voice-quality machinery, including moderation, crisis handling, watermarking, cloned-voice abuse scoring, and eval harnesses.

Implication:

  • Lilith should own contemplative teacher design language, meditation-tone policies, and spiritual-domain persona safety patterns.

Required canonical split#

Recommended durable ownership:

  • Oshun shell assistant = user-facing consumer UX and domain orchestration
  • Iris = persistent memory, conversation history, user memory controls, agent identity, and memory privacy
  • Psyche = real-time text/voice/avatar embodiment and live session runtime
  • Lilith = contemplative teacher/persona content, tone policy, and spiritual safety overlays
  • Sophia = grounding/citations when answers depend on evidence or source packs

Persona strategy recommendation#

Oshun should absolutely support a broad cast of agents and assistants, but not as an ungoverned pile of bespoke characters. The right model is a canonical persona registry with families such as:

  • contemplative teacher / meditation guide
  • philosopher / wisdom interlocutor
  • literature and textual tradition guide
  • self-discipline / Arete coach
  • Veritas explainer / reporter / source analyst
  • Nisaba research guide / comparative-text scholar
  • admin reviewer / moderator / support copilot

Each persona should have:

  • explicit role definition and capability envelope
  • disclosure policy
  • grounding policy
  • memory scope and retention rules
  • modality permissions
  • avatar style pack
  • evaluation set

The target should be "maximally natural and trustworthy" rather than "deceptive or undisclosed human impersonation." Stylized avatars can still feel premium, alive, and deeply present without crossing that line.

Major gaps#

  • no unified persona registry across Oshun, Iris, Psyche, and Lilith
  • no canonical avatar contract spanning still image, animated avatar, voice, turn-taking, and disclosure state
  • no product-wide memory consent model exposed consistently to end users
  • no clear boundary between consumer assistant personas and operator/admin copilots
  • no mature embodied assistant surface yet visible in Oshun mobile itself, even though richer mobile-oriented persona work exists in Lilith and related backing systems

16.4 Admin and operator surface audit#

The answer is yes: a unified Oshun product at this scope should have both customer-facing and admin-facing apps. The repo already shows why.

Existing operator/admin evidence#

  • Psyche already has a substantial admin web app in apps/psyche/admin, with navigation for dashboard, sessions, personas, knowledge base, tools, analytics, activity, and settings.
  • The broader repo also already contains dedicated admin-app patterns outside Oshun proper, including apps/aphrodite/admin and apps/aphrodite/admin-bi-dashboard. Those apps are not Oshun consumer backoffice surfaces, but they reinforce that the monorepo already treats admin products as first-class application types rather than mere hidden routes.
  • Sophia Workbench already includes review workflows, though parts still fall back to mock data.
  • Oshun web Studio already includes routes for review/approval workflows, audit/compliance surfaces, and observability/operational dashboards.
  • Lilith includes review/moderation-oriented components and reporting-system docs.
  • Oshun BFF exposes admin metrics endpoints documented in the on-call runbook.
  • Isis web surfaces also show review/security-operational patterns, for example output review loop utilities and region-retention compliance helpers.

What is missing#

  • no unified apps/oshun/admin
  • no dedicated Oshun admin mobile app
  • no single cross-domain operator console for:
    • trust and safety
    • content review
    • rights/licensing review
    • user support and appeals
    • billing/entitlements
    • DSAR/privacy operations
    • model and provider approval
    • persona/avatar approval
    • incident response and on-call

Recommendation#

Build Oshun Admin as a first-class product with:

  • web for full-fidelity operations
  • mobile for urgent queues, escalation, approvals, and incident response
  • shared evidence, provenance, and review objects so admin tools can operate across Tara, Arete, Veritas, Nyx, Nisaba, Sophia, Iris, Lilith, and Isis
  • internal operator copilots that assist but never silently finalize high-impact decisions

16.5 Agentic research and authoring audit#

Yemaya#

Yemaya looks like a valuable creative-authoring platform, but its runtime layer is still too prototype-shaped in several critical places.

Evidence:

  • apps/yemaya/api/src/services/agent-runtime.ts
    • explicitly describes an "In-memory agent registry + task execution runtime"
    • falls back to stub completions when no provider is configured
  • apps/yemaya/api/src/services/ai-generation.ts
    • stores availability cache, jobs, and prompt templates in memory maps
    • warns that AI generation is limited when providers are not configured

Design implication:

  • Yemaya should own authoring UX, composition flows, and project semantics.
  • It should not be the long-term durable orchestrator for product-critical agent runs, research reports, or generated asset history.

Lilith#

Lilith contains serious domain-specific logic, especially for meditation audio, TTS planning, safety, and voice quality. But some of its generation and RunPod integration services are still operationally immature.

Evidence:

  • apps/lilith/svc-ai/src/runpod-integration/runpod-integration-service.ts uses many in-memory maps and includes explicit "testing/simulation" comments in cache/download logic.
  • apps/lilith/svc-tts/src/voice-quality-validator.ts models strong quality certification concepts, but histories, verification records, and certificates are stored in memory maps.
  • apps/lilith/svc-ai/src/evals.ts already provides a meaningful eval harness with golden sets, adversarial prompts, and nightly-eval concepts.
  • apps/lilith/svc-tts/src/audio-watermarking-system.ts and apps/lilith/svc-tts/src/abuse-risk-scoring-engine.ts show materially real cloned-voice safety and traceability thinking.
  • apps/lilith/svc-media/src/app.ts and apps/lilith/svc-ai/src/image-moderation/* show that Lilith also has image/media moderation and watermark-oriented capability, not only audio safety.
  • docs/domains/lilith/extras/reference/safety-measures.md documents substantial moderation, crisis, PII, jailbreak, medical, and legal-advice guardrails.

Design implication:

  • Lilith should contribute domain-specific audio quality policy, meditation scripting, and voice safety logic.
  • Lilith is stronger as a policy/safety/tone layer than the earlier report version implied.
  • Durable workflow execution, provenance, and provider orchestration should be externalized into shared platform services.

Aphrodite#

The Aphrodite research-agents package is rich in ambition but still reads as simulation-heavy.

Evidence:

  • libs/aphrodite/research-agents/src/agent-engine.ts includes simulated document parsing behavior, pseudo-random simulation helpers, and large in-memory state maps.

Design implication:

  • Aphrodite is useful as an experimentation and modeling layer.
  • It should not be treated as the final authoritative research runtime for high-trust Oshun outputs without a major productionization pass.

Psyche#

Psyche is the most serious-looking live assistant stack in the repo.

Evidence:

  • services/psyche/README.md describes a multi-service platform spanning avatar, voice, behavior, perception, knowledge base, persona, learning, and orchestration.
  • services/psyche/orchestrator/README.md describes Redis-backed state, resource allocation, semantic caching, multi-party coordination, and production-oriented orchestration concepts.
  • Tavus-related services and libraries indicate substantial avatar/persona and conferencing integration work, rather than a toy assistant prototype.
  • Some local orchestrator/session state code still exposes in-memory storage patterns, so Psyche should be treated as the strongest embodiment candidate, not as fully production-hardened by default.

Design implication:

  • Psyche should likely own real-time multimodal assistant/session orchestration.
  • Oshun should avoid recreating that orchestration logic independently in lighter-weight TypeScript services unless there is a strong reason.
  • Psyche itself still needs the same durability/replay/review hardening expected of the rest of the stack.

16.6 Image, audio, and video generation audit#

ComfyUI#

Official ComfyUI docs now expose:

  • Cloud API operations
  • object introspection
  • input upload flows
  • workflow execution
  • custom node development and registry-style extension patterns

Local Oshun docs already frame ComfyUI as a multi-provider workflow layer across RunComfy, RunPod, Modal, Replicate, and Vast.ai.

Design implication:

  • Treat ComfyUI as a workflow IR and execution adapter layer.
  • Do not let product teams treat raw ComfyUI graphs as the product contract.
  • Versioned workflow templates and policy should sit above ComfyUI.

RunPod#

Official RunPod docs now emphasize:

  • serverless endpoints with scaling, timeouts, and GPU selection
  • queue-based and load-balancing endpoint modes
  • cached models for faster cold starts and lower cost
  • console, REST API, SDK, and CLI access paths

The local repo already has unusually strong RunPod operational thinking in docs/infrastructure/runpod-integration.md:

  • endpoint classes
  • warm/cold policy
  • pay-per-second attribution
  • per-endpoint SLOs
  • anomaly detection
  • automated canary strategy
  • fleet topology and economics

Design implication:

  • Oshun should centralize RunPod policy and execution behind one platform boundary.
  • Product domains should not each negotiate their own GPU/control-plane rules.

ElevenLabs#

Official ElevenLabs docs now expose a broader capability family than legacy TTS thinking suggests:

  • text to speech
  • speech to text
  • music
  • text to dialogue
  • image and video capabilities in the docs navigation
  • voice changer, voice isolator, dubbing, sound effects, voice library, and voice agents

For TTS specifically, current official docs highlight:

  • Flash v2.5 for ultra-low latency
  • Multilingual v2 for long-form stability
  • Eleven v3 for more expressive dialogue-oriented synthesis
  • thousands of voices via library, plus cloning and voice design
  • text-to-dialogue as a strong authored/multi-speaker capability, but not as the primary path for low-latency conversational agents

Design implication:

  • ElevenLabs should be treated as one provider inside a broader Oshun voice platform, not as the platform itself.
  • For real-time assistants, Oshun should prefer low-latency conversational stacks; for authored dialogue, meditations, or premium scripted experiences, richer offline or batch synthesis modes are appropriate.
  • Oshun still needs its own voice policy, prompt policy, brand-voice governance, pronunciation/dictionary controls, and evaluation harness.

Civitai#

The repo's Civitai posture is broader than the earlier report credited.

Existing evidence:

  • docs/integrations/civitai/* documents model discovery, generation, training, model-management, and operational best practices.
  • libs/isis/ai-providers/src/providers/civitai/* includes provider, link, recommendation, comparison, and usage-control surfaces.
  • apps/lilith/svc-ai/src/providers/civitai/* includes model management, discovery, and sync-style logic.
  • apps/oshun/web/src/app/studio/isis/civitai-integration/page.tsx exposes an Oshun Studio route for Civitai workflows.

Corrections and gaps:

  • The local repo includes simulation and in-memory patterns in key Civitai provider paths, including explicit "In production, this would call Civitai API" comments and mock/simulation layers.
  • Public official documentation is clearest around discovery/library-style model APIs and model-sharing workflows, not around treating Civitai as the canonical execution runtime for Oshun.
  • The strongest architecture is therefore:
    • Civitai = external model intelligence, discovery, metadata, preview assets, and community compatibility signal
    • Isis = canonical model registry, staging, provenance, approval, and runtime control plane
    • ComfyUI/RunPod = owned workflow execution substrates

Recommended Civitai integration model:

  • ingest Civitai models, versions, preview images, creator metadata, hashes, licenses, NSFW/content labels, and recommended parameters into Isis ModelRegistry
  • store approved preview assets and model cards inside Oshun-controlled storage
  • attach tested workflow presets, VRAM/runtime guidance, safety class, and compatibility recommendations
  • require review/approval before a model becomes production-eligible
  • generate staging recipes for ComfyUI/RunPod rather than assuming Civitai is the runtime of record

Isis#

Among the repo's generation systems, Isis is the clearest candidate for the canonical control plane.

Strengths:

  • explicit generation API, GPU worker, workflow registry, and output registry
  • PostgreSQL/Prisma schema with lineage/provenance
  • Redis queue model
  • provider adapter layer
  • workflow versioning
  • generation-type coverage across image, video, audio, and 3D
  • ADRs for RunPod/serverless strategy

Recommendation:

  • Isis should become the mandatory workflow/provenance/control plane for image/audio/video/3D generation across the Oshun ecosystem.

16.7 Cross-cutting quality gaps#

Across the backing systems, the biggest gaps are:

  • no single product-wide capability registry for grounding, generation, orchestration, evaluation, and publication
  • no universal provenance standard spanning:
    • sources and citations
    • generated outputs
    • model/provider/workflow versions
    • agent steps and tool calls
  • no single evaluation layer for:
    • groundedness and citation integrity
    • spiritual/meditative tone quality
    • voice quality and safety
    • image/video aesthetic and consistency quality
  • no canonical persona registry or avatar contract that spans memory, disclosure, embodiment, and moderation state
  • no product-wide consent/watermark/review gate that consistently covers cloned voices, embodied avatars, generated research reports, and generated media
  • no clean separation between:
    • demo/simulation packages
    • production runtimes
    • domain logic libraries
    • platform control planes
  • inconsistent persistence: too many quality-critical workflows still keep state in process memory
  • no universal editorial lifecycle for content, reports, assets, and generated experiences
  • no unified admin operator product for high-priority moderation, support, compliance, model-ops, and incident-response workflows

16.8 Significant rearchitecture and refactor recommendations#

These are the changes I would consider materially important, not optional polish:

  1. Establish a shared Oshun AI platform contract. It should define canonical capability interfaces such as:
    • ResearchGroundingService
    • CitationService
    • KnowledgeGraphService
    • AssistantMemoryService
    • PersonaRegistryService
    • AvatarRuntimeService
    • AgentRuntimeService
    • GenerationWorkflowService
    • VoiceSynthesisService
    • EvaluationService
    • PublishingService
  2. Make Sophia the canonical grounding/citation substrate. Veritas, Nisaba, assistant, and studio should call Sophia-backed services instead of each inventing their own retrieval and citation logic.
  3. Make Iris the canonical assistant memory and identity substrate, and make Psyche the canonical real-time embodiment/session substrate. The Oshun shell should stop pretending its local assistant runtime is the whole assistant platform.
  4. Make Isis the canonical generation control plane. Domains can request assets; they should not directly own provider-specific orchestration unless absolutely necessary.
  5. Treat Civitai as a model-intelligence and sourcing layer, not as the platform-of-record runtime. Normalize Civitai metadata into Isis, then stage approved models onto owned execution substrates.
  6. Build a canonical persona registry and avatar contract. Every persona should have a durable identity, memory scope, safety policy, disclosure state, modality permissions, evaluation set, and approved embodiment pack.
  7. Build Oshun Admin Web and Oshun Admin Mobile as first-class products. Operator tooling is already fragmented enough that this is now architecture, not polish.
  8. Replace in-memory runtime state in quality-critical services with durable storage and queues. Anything that matters for user trust, billing, provenance, or continuation should survive process restarts.
  9. Split prototype/simulation codepaths from production services. A clear repo distinction is needed between:
    • simulation/demo/reference implementations
    • production-grade service implementations
  10. Add a universal trace and replay model for agents and generation jobs. Every meaningful run should preserve:
  • prompt/workflow version
  • tools called
  • sources retrieved
  • outputs produced
  • evaluations run
  • user-facing artifact IDs
  1. Create a unified evaluation pipeline. Minimum gates should cover:
  • text groundedness
  • citation integrity
  • voice quality and pronunciation
  • clone/abuse safety
  • image/video quality and policy checks
  • domain-specific contemplative-tone checks where relevant
  1. Add a unified consent, disclosure, and watermark policy for embodied AI. Highest-end assistants should feel natural, not deceptive. Users should always know when they are speaking with AI, what memory is active, and what content is synthetic or voice-cloned.
  2. Create one product graph linking ritual, research, generation, and user memory. Shared objects should connect concepts, sources, passages, claims, assets, rituals, projects, and saved outputs.
  3. Introduce an editorial/publishing lifecycle. Research reports, generated meditations, voice assets, study packs, and domain explainers should support review states such as:
  • draft
  • evaluated
  • reviewed
  • published
  • deprecated
  1. Clarify platform ownership and stop overlapping orchestration. The likely ownership split is:
    • Sophia: grounding and citation
    • Iris: assistant memory and privacy boundary
    • Psyche: live assistant sessions and embodiment
    • Isis: generation and model control plane
    • Yemaya: authoring UX/composition
    • Lilith: meditation/voice domain logic and contemplative safety/tone
    • Aphrodite: experimental/autonomous research agents until hardened

16.9 What this means for Sophia, Nisaba, embodied assistants, and admin apps#

This audit changes the answer to "how about Sophia?" and also sharpens the answer to "what about embodied assistants and admin products?"

  • Nisaba should be added as a user-facing domain in Oshun web and mobile.
  • Sophia should be added to the report, roadmap, and platform architecture as a first-class system, but not necessarily as a default shell card.
  • The web app should likely expose Sophia most directly through:
    • pro research/workbench surfaces
    • Studio/Hathor grounding tools
    • Veritas/Nisaba evidence views
  • The mobile app should consume Sophia mostly indirectly through:
    • grounded answers
    • saved source packs
    • note/citation continuation
    • explainers and compare views
  • Oshun should also gain:
    • an admin web app for review, moderation, support, model ops, persona ops, and governance
    • an admin mobile app for urgent operational workflows
    • a premium embodied assistant layer that uses Iris for memory, Psyche for real-time text/voice/avatar interaction, Lilith for contemplative teacher logic, and Sophia for evidence-backed responses when needed

In short:

  • Nisaba = a top-level depth domain
  • Sophia = the substrate that makes that depth trustworthy
  • Iris + Psyche + Lilith = the stack that can make Oshun's assistants feel premium, coherent, and natural
  • Oshun Admin = the product required to operate all of the above responsibly

17. Sequenced Implementation Roadmap#

Phase 0: Product alignment#

Decide and document:

  • Tara-led product narrative
  • Nisaba role definition
  • Sophia product-vs-platform decision
  • admin product strategy
  • embodied assistant/persona strategy
  • platform capability ownership across Sophia/Iris/Isis/Yemaya/Lilith/Psyche/Aphrodite
  • domain role model
  • platform role model
  • definition of "integrated"

Output:

  • updated product brief
  • updated domain strategy docs
  • platform capability ownership map
  • admin surface strategy
  • persona and memory governance policy

Phase 1: Shell contract expansion#

  • add nisaba to canonical domain enums and lists
  • add registry metadata
  • add token/icon/domain label support
  • update route validators
  • update auth, analytics, assistant, notifications, BFF types
  • define shared evidence/citation object typing for grounded flows
  • define shared review/provenance object typing for admin surfaces
  • define canonical assistant memory/persona contracts

Output:

  • Nisaba recognized system-wide

Phase 2: Basic launch surfaces#

  • web Nisaba landing surface
  • mobile Nisaba companion launch surface
  • search/filter/library/activity inclusion
  • assistant awareness
  • minimal grounded source/citation views backed by Sophia
  • initial Oshun Admin web information architecture
  • initial Iris-backed shell memory handoff

Output:

  • Nisaba visible and usable inside Oshun

Phase 3: Nisaba MVP utility#

  • web:
    • reader
    • compare
    • concepts
    • saved annotations
  • mobile:
    • daily passage
    • compare
    • notes/highlights
    • reminders

Output:

  • Nisaba is no longer just visible; it has daily utility

Phase 4: Cross-domain activation#

  • Tara x Nisaba contemplative text journeys
  • Arete x Nisaba study habit and journal loops
  • Veritas x Nisaba provenance and primary-source context
  • Nyx x Nisaba cosmology and calendar overlays

Output:

  • the fifth domain increases cohesion instead of fragmentation

Phase 5: Deep-work maturity#

  • projects/workspaces
  • advanced comparative tools
  • manuscript/edition workflows
  • exports and standards

Output:

  • Nisaba becomes Oshun's signature depth moat

Phase 6: Admin product unification#

  • build apps/oshun/admin web shell
  • unify review, moderation, support, entitlement, privacy, and model-ops queues
  • expose shared audit/provenance/evidence objects
  • create admin mobile companion for urgent queues, escalations, and on-call

Output:

  • Oshun gains an operator product that can run the unified ecosystem

Phase 7: Embodied assistant and persona platform#

  • wire Oshun shell assistant to Iris memory
  • route real-time multimodal sessions through Psyche
  • establish canonical persona registry and avatar contract
  • launch premium persona families across consumer and admin contexts
  • add explicit consent, disclosure, memory controls, and synthetic-media signals

Output:

  • Oshun assistants become coherent, premium, and governable across text, voice, and avatar modes

Phase 8: Model intelligence and media hardening#

  • normalize Civitai discovery/metadata into Isis model registry
  • attach tested presets, safety labels, licenses, and staging recipes
  • centralize ComfyUI/RunPod execution through Isis
  • add universal release gates for grounded reports, voices, avatars, and media

Output:

  • Oshun media and model operations become reviewable, portable, and production-grade

18. Risks and Anti-Patterns#

18.1 Biggest risks#

  • adding Nisaba only as another card
  • treating Sophia as optional instead of as the grounding substrate
  • letting Arete drift into generic self-optimization
  • keeping Tara siloed as wellness content
  • treating web and mobile parity as literal duplication
  • using cross-domain recommendations without a clear conceptual model
  • relying on prototype-shaped agent/generation runtimes for quality-critical production behavior
  • letting embodied assistants become deceptive, unsafe, or memory-invasive
  • shipping admin/operator responsibilities as scattered workbenches instead of one governed product
  • over-trusting external provider capabilities or docs without normalizing them into Oshun-owned contracts and review gates
  • expanding breadth faster than meaning

18.2 Anti-patterns to avoid#

  • "super-app" as a pile of tabs
  • forced symmetry between platforms
  • shallow cross-links with no real context
  • separate saves, separate histories, separate semantics
  • scholarship isolated from lived practice
  • ad hoc provider integration from every domain instead of shared control planes
  • black-box AI outputs with no provenance or evaluation trail
  • "indistinguishable from a person" as a product objective without mandatory AI disclosure, consent, and memory controls
  • treating admin mobile as a full clone of admin web instead of a high-trust, urgent-workflows companion

Measure cohesion, not just feature count.

Recommended product metrics:

  • percent of weekly active users engaging with 2+ domains
  • percent of weekly active users completing a cross-domain ritual
  • share of Tara sessions initiated from non-Tara contexts
  • share of Veritas sessions that lead to a save, note, or follow-up action
  • share of Nyx events that lead to ritual, observation, or reflection actions
  • share of Nisaba usage originating from another domain
  • number of collections containing mixed-domain items
  • assistant requests that span multiple domains
  • "continue" rate across devices and platforms
  • grounded-answer citation coverage rate
  • citation integrity pass rate
  • percent of generation runs with full provenance bundles
  • percent of agent runs with durable traces and replayability
  • audio/image/video first-pass evaluation acceptance rate
  • share of assistant turns that use canonical Iris memory and Sophia grounding when applicable
  • percentage of embodied sessions with valid disclosure, consent, and memory scope recorded
  • percentage of cloned-voice or avatar sessions with watermark/provenance bundle attached

Recommended Nisaba-specific metrics:

  • daily passage open rate
  • compare-translation usage
  • save/highlight/annotation conversion
  • concept-thread engagement
  • web project/workspace continuation
  • cross-domain attachment rate from Nisaba to Tara/Arete/Veritas/Nyx

Recommended Sophia/platform metrics:

  • share of Veritas explanations backed by Sophia citations
  • share of assistant answers exposing inspectable sources
  • notebook/report export completion rate
  • source-pack save rate
  • evidence-view open rate after claim inspection
  • percentage of product surfaces using canonical grounding and generation services rather than bespoke adapters
  • moderation appeal overturn rate
  • human-review SLA by queue type
  • admin mobile acknowledgment time for urgent queues
  • Civitai import success rate into Isis model registry
  • percentage of Civitai-sourced models with completed license/content/hash review
  • sync freshness for external model metadata and recommendations

20. Final Recommendations#

Recommendation 1#

Reframe Oshun immediately around a Tara-led contemplative intelligence model.

Recommendation 2#

Treat Nisaba as the next strategic domain, but integrate it as a depth layer, not a silo.

Recommendation 3#

Treat Sophia as the canonical grounding and citation substrate, not just another possible tab.

Recommendation 4#

Introduce shared cross-domain objects, evidence objects, and rituals before adding too much more surface area.

Recommendation 5#

Promote Isis to the canonical generation control plane for image/audio/video workflows and stop ad hoc provider sprawl.

Recommendation 6#

Build Oshun Admin Web and Oshun Admin Mobile as first-class products instead of continuing with fragmented workbenches.

Recommendation 7#

Make the assistant stack explicit:

  • Iris for memory and assistant continuity
  • Psyche for real-time embodiment
  • Lilith for contemplative teacher logic and safety
  • Sophia for grounding
  • Oshun shell for UX orchestration

Recommendation 8#

Create a canonical persona registry and avatar contract before scaling the cast of teachers, reporters, philosophers, coaches, and other assistants.

Recommendation 9#

Make platform role clarity explicit:

  • parity where needed
  • web depth where appropriate
  • mobile companion loops where stronger

Recommendation 10#

Use the assistant, collections, concept graph, and evidence graph as the primary cohesion systems.

Recommendation 11#

Replace prototype-shaped in-memory agent/generation runtimes in quality-critical paths with durable orchestration, provenance, and evaluation.

Recommendation 12#

Treat Civitai as a discovery/intelligence and sourcing layer. Normalize all externally sourced model and asset metadata into Isis before operational use.

Recommendation 13#

Do not pursue "indistinguishable from a person" as a deceptive goal. Pursue highest-end natural interaction with explicit disclosure, consent, memory controls, reviewability, and provenance.


21. Appendix A: Key Repository Evidence#

Primary shell/domain contract files:

  • libs/oshun/navigation/src/routes.ts
  • libs/oshun/domain-registry/src/index.ts
  • libs/oshun/auth/src/types.ts
  • libs/oshun/shell-assistant/src/types.ts
  • apps/oshun/web/src/navigation/routes.ts
  • apps/oshun/mobile/src/navigation/routes.ts
  • apps/oshun/bff/src/routes/feed-helpers.ts

Web launch and domain experience files:

  • apps/oshun/web/src/app/domains/[domainId]/page.tsx
  • apps/oshun/web/src/components/DomainRouteExperience.tsx

Mobile launch and shell files:

  • apps/oshun/mobile/app/domain/[domainId].tsx
  • apps/oshun/mobile/app/(tabs)/index.tsx
  • apps/oshun/mobile/src/components/DomainSwitcherEntry.tsx

Strategic documentation:

  • docs/releases/v1/specs/product-brief.md
  • docs/releases/v1/scope/mvp-scope.md
  • docs/reference/api-contracts.md
  • docs/reference/data-ownership-matrix.md
  • docs/reference/monitoring.md
  • docs/releases/v1/specs/studio-domain-dependency-and-data-contract-map.md
  • docs/migration/yemaya-migration-guide.md

Nisaba-specific documentation and libraries:

  • docs/releases/v1/domains/nisaba/README.md
  • docs/releases/v1/domains/nisaba/feature-inventory.md
  • docs/releases/v1/domains/nisaba/implementation-map.md
  • libs/nisaba/client/src/react-hooks.ts
  • libs/nisaba/workspace/src/research-environment.ts

Sophia and grounding evidence:

  • docs/releases/v1/domains/sophia/README.md
  • docs/releases/v1/domains/sophia/implementation-map.md
  • libs/openapi/src/specs/sophia/sophia-api.yaml
  • libs/proto/src/sophia/sophia.proto
  • docs/domains/sophia/architecture.md
  • docs/domains/sophia/database.md
  • apps/sophia/knowledge-graph/src/main.ts
  • apps/sophia/workbench/src/api/services/review-service.ts
  • libs/sophia/theory/src/citations/citation-manager.ts
  • libs/sophia/vectordb/src/index.ts

Iris, assistant, and memory evidence:

  • docs/releases/v1/domains/iris/README.md
  • docs/domains/iris/features.md
  • libs/contracts/iris/src/conversation/index.ts
  • libs/contracts/iris/src/memory/index.ts
  • libs/oshun/shell-assistant/src/assistant-engine.ts
  • apps/oshun/web/src/components/assistant/AssistantPanel.tsx

Admin and operator evidence:

  • apps/psyche/admin/src/components/layout/sidebar.tsx
  • apps/aphrodite/admin/src/routes/moderation.ts
  • apps/aphrodite/admin-bi-dashboard/src/routes/compliance.ts
  • docs/domains/psyche/admin-dashboard-guide.md
  • apps/oshun/web/src/app/studio/review-approval-workflows/page.tsx
  • apps/oshun/web/src/app/studio/audit-compliance-surfaces/page.tsx
  • apps/oshun/web/src/app/studio/observability-operational-dashboards/page.tsx
  • docs/releases/v1/runbooks/on-call-runbook.md
  • apps/isis/web/src/pages/output-review-loop-utils.ts
  • apps/isis/web/src/security/region-retention-compliance-utils.ts
  • docs/domains/lilith/extras/reference/reporting-system.md
  • docs/domains/lilith/extras/reference/safety-measures.md
  • docs/domains/lilith/extras/ai-teacher-personality-registry.md
  • apps/lilith/mobile/*

Agentic/generation/media systems evidence:

  • docs/domains/isis/README.md
  • docs/infrastructure/runpod-integration.md
  • docs/integrations/civitai/README.md
  • docs/integrations/civitai/api.md
  • docs/integrations/civitai/model-management.md
  • docs/integrations/civitai/training.md
  • apps/oshun/web/src/app/studio/isis/civitai-integration/page.tsx
  • libs/isis/ai-providers/src/providers/civitai/civitai-link-provider.ts
  • docs/integrations/comfyui/README.md
  • docs/integrations/elevenlabs/README.md
  • openapi/isis/openapi.yaml
  • apps/yemaya/api/src/services/agent-runtime.ts
  • apps/yemaya/api/src/services/ai-generation.ts
  • apps/lilith/svc-ai/src/advanced/agentic-research.ts
  • apps/lilith/svc-ai/src/evals.ts
  • apps/lilith/svc-ai/src/image-moderation/service.ts
  • apps/lilith/svc-ai/src/runpod-integration/runpod-integration-service.ts
  • apps/lilith/svc-tts/src/audio-watermarking-system.ts
  • apps/lilith/svc-tts/src/abuse-risk-scoring-engine.ts
  • apps/lilith/svc-tts/src/voice-quality-validator.ts
  • apps/lilith/svc-media/src/app.ts
  • services/psyche/README.md
  • services/psyche/orchestrator/README.md
  • libs/aphrodite/research-agents/src/agent-engine.ts

Meditation / contemplative apps:

  • Headspace Move Mode: https://help.headspace.com/hc/en-us/articles/360043063714-What-is-Move-Mode
  • Calm Premium feature overview: https://support.calm.com/hc/en-us/articles/360008536834-Calm-Premium-vs-Free-Features-Content-List-Benefits
  • Insight Timer live events support article: https://help.insighttimer.com/support/solutions/articles/67000665284-applying-for-live-events-creating-a-live-event-proposal

Habits / productivity / gamification:

  • Fabulous product site: https://www.thefabulous.co/
  • Todoist Karma: https://www.todoist.com/de/karma
  • Duolingo streak celebration article: https://blog.duolingo.com/de/streak-feiern/

News / truth / research organization:

  • Google News Full Coverage in Search: https://blog.google/products-and-platforms/products/news/get-full-news-story-full-coverage-search/
  • Ground News media bias page: https://ground.news/media-bias
  • Feedly navigation and AI feed docs: https://docs.feedly.com/article/801-how-to-navigate-through-feedly

Astronomy / exploration:

  • Stellarium official site page: https://stellarium.org/en/landscapes-south-america.html
  • SkySafari LiveSky data management and sync: https://www.skysafariastronomy.com/livesky-data-management-and-cloud-sync-service.html
  • NASA JPL Apps / Eyes on the Solar System: https://www.jpl.nasa.gov/apps

Textual scholarship / libraries / manuscript tools:

  • Sefaria interconnected texts help article: https://help.sefaria.org/hc/en-us/articles/18613227644316-How-to-Find-Interconnected-Texts
  • Logos lemma study workflow: https://support.logos.com/hc/en-us/articles/28716717196813-Study-a-Lemma-with-Logos
  • Transkribus handwriting-to-text platform page: https://www.transkribus.org/handwriting-to-text
  • Tropy metadata docs: https://docs.tropy.org/before-you-begin/metadata

Research / grounding / agentic systems:

  • NotebookLM official blog post on source grounding and product direction: https://blog.google/innovation-and-ai/products/notebooklm-new-features-availability/
  • Perplexity Deep Research official announcement: https://www.perplexity.ai/hub/blog/introducing-perplexity-deep-research
  • Elicit systematic reviews help article: https://support.elicit.com/en/articles/7927169
  • Elicit research reports help article: https://support.elicit.com/en/articles/4168449
  • Semantic Scholar open data and API FAQ: https://www.semanticscholar.org/faq/open-data

Generation / media infrastructure:

  • Civitai official GitHub wiki home / REST API reference entry: https://github.com/civitai/civitai/wiki
  • ComfyUI Cloud API reference: https://docs.comfy.org/development/cloud/api-reference
  • ComfyUI custom nodes documentation: https://docs.comfy.org/development/core-concepts/custom-nodes
  • RunPod endpoint overview: https://docs.runpod.io/serverless/endpoints/overview
  • RunPod workflow/connectivity overview: https://docs.runpod.io/get-started/connect-to-runpod
  • RunPod serverless workers overview: https://docs.runpod.io/serverless/workers/overview
  • ElevenLabs text-to-speech capability docs: https://elevenlabs.io/docs/overview/capabilities/text-to-speech
  • ElevenLabs speech-to-text help/docs entry: https://help.elevenlabs.io/hc/en-us/articles/33053029255697-What-is-Speech-to-Text
  • ElevenLabs text-to-dialogue capability docs: https://elevenlabs.io/docs/overview/capabilities/text-to-dialogue

23. Closing Position#

OSHUN should not merely become "the app that contains Tara, Veritas, Nyx, Arete, and Nisaba."

It should become the product where:

  • Tara gives the user center,
  • Arete gives the user practice,
  • Veritas gives the user clarity,
  • Nyx gives the user wonder,
  • Nisaba gives the user depth,
  • Sophia quietly makes the whole system grounded,
  • Iris and Psyche make assistance continuous and alive,
  • and Oshun Admin makes the whole ecosystem governable.

That is the cohesive product this repository is already hinting at, but has not yet fully implemented.