Eve library

Isis Database Schema

Guide

Revision 1

On this page

Isis is the Generative AI Factory domain. Its database stores generation jobs, GPU worker assignments, workflow definitions, generated output files with full provenance tracking, model/checkpoint registry, asset lineage graphs, and retention policies.

Database: PostgreSQL ORM: Prisma Schema Location: libs/isis/database/prisma/schema.prisma Environment Variable: ISIS_DATABASE_URL Generated Client: libs/isis/database/src/generated/client

ER Diagram#

text
+---------------------+       +---------------------+
|   ModelRegistry     |       |      Workflow        |
|---------------------|       |---------------------|
| id (PK)             |       | id (PK)             |
| name (UQ w/version) |       | name (UQ w/ownerId) |
| type                |       | engine              |
| format              |       | category            |
| version             |       | status              |
| storagePath         |       | visibility          |
| sizeBytes           |       | currentVersion      |
| hashAlgorithm       |       | definition          |
| hashValue           |       | parameters          |
| triggerWords[]      |       | runCount            |
| tags[]              |       | starCount           |
| isPublic            |       | ownerId             |
+--------+------------+       +--------+------------+
         |                             |
         |1                            |1
         |                             |
         |*                            |*
+--------+------------+       +--------+------------+
|   GenerationJob     |       |  WorkflowVersion    |
|---------------------|       |---------------------|
| id (PK)             |       | id (PK)             |
| type                |       | workflowId (FK)     |
| status              |       | version (UQ w/wf)   |
| priority            |       | changelog?          |
| prompt?             |       | definition          |
| negativePrompt?     |       | deprecated          |
| modelId (FK)?       |       +---------------------+
| workflowId (FK)?    |
| parameters          |       +---------------------+
| gpuType?            |       |   WorkflowStar      |
| gpuTimeSeconds?     |       |---------------------|
| costEstimate?       |       | id (PK)             |
| userId              |       | workflowId (FK)     |
| projectId?          |       | userId (UQ w/wf)    |
+--------+------------+       +---------------------+
         |
         |1                   +---------------------+
         |                    |  WorkflowTemplate   |
         |*                   |---------------------|
+--------+------------+       | id (PK)             |
|   GeneratedOutput   |       | name                |
|---------------------|       | engine              |
| id (PK)             |       | category            |
| name                |       | definition          |
| type                |       | featured            |
| mimeType            |       | usageCount          |
| size                |       +---------------------+
| hashAlgorithm       |
| hashValue           |
| storageBucket       |       +---------------------+
| storageKey (UQ w/b) |       |    Provenance       |
| storageTier         |       |---------------------|
| status              |       | id (PK)             |
| width?              |       | outputId (FK, UQ)   |
| height?             |       | jobId (FK, UQ)?     |
| duration?           |       | workflowInfo?       |
| jobId (FK)?         |       | modelInfo?          |
| workflowId (FK)?    |       | lorasUsed           |
| ownerId             |       | parameters          |
| tags[]              |       | inputs              |
+--------+------------+       | seed?               |
         |                    | gpu?                |
         |1                   | provider?           |
         |                    | cost?               |
         |*         1         +---------------------+
+--------+------------+
|    LineageEdge      |
|---------------------|
| id (PK)             |
| sourceId (FK)       | ---> GeneratedOutput (LineageSource)
| targetId (FK)       | ---> GeneratedOutput (LineageTarget)
| type                |
| transformation?     |
| weight?             |
+---------------------+

+---------------------+       +---------------------+
|     GpuWorker       |       |  RetentionPolicy    |
|---------------------|       |---------------------|
| id (PK)             |       | id (PK)             |
| name                |       | name (UQ)           |
| hostname (UQ)       |       | hotRetentionDays    |
| gpuType             |       | warmRetentionDays   |
| gpuMemoryMb         |       | coldRetentionDays   |
| gpuCount            |       | totalRetentionDays  |
| supportedTypes[]    |       | outputTypes[]       |
| supportedEngines[]  |       | isActive            |
| status              |       | organizationId?     |
| lastHeartbeat?      |       +---------------------+
| totalJobsProcessed  |
+---------------------+       +---------------------+
                              |     AuditLog        |
                              |---------------------|
                              | id (PK)             |
                              | action              |
                              | entityType          |
                              | entityId            |
                              | previousState?      |
                              | newState?           |
                              | userId?             |
                              | createdAt           |
                              +---------------------+

Models#

GenerationJob#

The core work unit. Each generation job represents a single AI generation request (text-to-image, image-to-video, 3D generation, etc.) with full lifecycle tracking.

Field Type Default Description
id String (cuid) auto Primary key
type GenerationType - Generation type
status JobStatus PENDING Job lifecycle status
priority JobPriority NORMAL Queue priority
progress Int 0 Progress percentage (0-100)
prompt String? - Text prompt
negativePrompt String? - Negative prompt
inputUrl String? - Input file URL
modelId String? - FK to ModelRegistry
workflowId String? - FK to Workflow
parameters Json {} Generation parameters
outputData Json? - Output metadata
previewUrl String? - Preview image URL
errorMessage String? - Error message
errorCode String? - Error code
retryCount Int 0 Retry count
maxRetries Int 3 Max retries
workerId String? - Assigned GPU worker
queueName String? - Queue name
queuePosition Int? - Queue position
callbackUrl String? - Webhook callback URL
metadata Json {} Arbitrary metadata
gpuType String? - GPU type used
gpuTimeSeconds Int? - GPU time consumed
tokensUsed Int? - LLM tokens consumed
costEstimate Float? - Estimated cost
userId String - Owner user ID
organizationId String? - Owner organization
projectId String? - Project reference
createdAt DateTime now() Created
updatedAt DateTime auto Updated
startedAt DateTime? - Processing start
completedAt DateTime? - Processing end
estimatedCompletionAt DateTime? - ETA
deletedAt DateTime? - Soft delete

Relations: model (ModelRegistry?), workflow (Workflow?), outputs (GeneratedOutput[]), provenance (Provenance?) Table: generation_jobs

Workflow#

Reusable workflow definitions for ComfyUI, Blender, Unreal, Godot, or custom pipelines. Includes versioning, visibility controls, and community statistics.

Field Type Default Description
id String auto Primary key
name String - Workflow name
description String? - Description
engine WorkflowEngine - Engine type
category WorkflowCategory - Category
status WorkflowStatus DRAFT Lifecycle status
visibility WorkflowVisibility PRIVATE Access level
currentVersion String "1.0.0" Current version
totalVersions Int 1 Total version count
parameters Json [] Workflow parameters schema
outputs Json [] Expected outputs schema
requirements Json {} Hardware/software requirements
definition Json {} Workflow definition (nodes/edges)
previewImage String? - Preview image URL
tags String[] [] Tags
runCount Int 0 Total runs
successfulRuns Int 0 Successful runs
avgExecutionTime Float? - Average execution time
starCount Int 0 Stars/likes
downloadCount Int 0 Downloads
author String - Author name
ownerId String - Owner user ID
organizationId String? - Organization
searchVector tsvector? - Full-text search

Unique: [ownerId, name] Relations: versions (WorkflowVersion[]), jobs (GenerationJob[]), outputs (GeneratedOutput[]), stars (WorkflowStar[]) Table: workflows

WorkflowVersion#

Immutable snapshot of a workflow definition at a specific version point.

Field Type Default Description
id String auto Primary key
workflowId String - FK to Workflow
version String - Semver string
changelog String? - Version changelog
definition Json - Workflow definition snapshot
parameters Json [] Parameters snapshot
requirements Json {} Requirements snapshot
deprecated Boolean false Whether deprecated
author String - Version author
metadata Json {} Metadata
createdAt DateTime now() Created

Unique: [workflowId, version] Table: workflow_versions

WorkflowStar#

User star/like on a workflow (many-to-many between users and workflows).

Unique: [workflowId, userId] Table: workflow_stars

WorkflowTemplate#

Curated workflow templates for quick-start creation.

Field Type Default Description
id String auto Primary key
name String - Template name
description String? - Description
engine WorkflowEngine - Engine type
category WorkflowCategory - Category
definition Json - Workflow definition
parameters Json [] Default parameters
previewImage String? - Preview URL
tags String[] [] Tags
featured Boolean false Whether featured
usageCount Int 0 Usage count

Table: workflow_templates

GeneratedOutput#

Every file produced by a generation job. Tracks full storage details, media dimensions, access counts, and participates in the lineage graph.

Field Type Default Description
id String auto Primary key
name String - Output filename
description String? - Description
type OutputFileType - File type category
mimeType String - MIME type
size BigInt - Size in bytes
hashAlgorithm HashAlgorithm SHA256 Hash algorithm
hashValue String - File hash
storageBucket String - Object storage bucket
storageKey String - Object storage key
storageTier StorageTier HOT Storage tier
storageRegion String? - Storage region
status OutputStatus PENDING Availability status
width Int? - Width (pixels)
height Int? - Height (pixels)
depth Int? - Depth (for 3D)
duration Float? - Duration (seconds, for AV)
frameCount Int? - Frame count (for video)
jobId String? - FK to GenerationJob
workflowId String? - FK to Workflow
ownerId String - Owner user ID
organizationId String? - Organization
projectId String? - Project reference
tags String[] [] Tags
metadata Json {} Metadata
accessCount Int 0 Access counter
lastAccessedAt DateTime? - Last accessed

Unique: [storageBucket, storageKey] Relations: job, workflow, provenance, parentEdges (LineageEdge[]), childEdges (LineageEdge[]) Table: generated_outputs

LineageEdge#

Directed edge in the output lineage graph. Tracks how one output was derived from another (upscaled, composed, refined, etc.).

Field Type Default Description
id String auto Primary key
sourceId String - FK to source GeneratedOutput
targetId String - FK to target GeneratedOutput
type LineageEdgeType - Edge type
transformation String? - Transformation description
weight Float? 1.0 Edge weight

Unique: [sourceId, targetId] Table: lineage_edges

Provenance#

Full generation provenance for an output, linking the output to the job, workflow, model, LoRAs, seed, and hardware used.

Field Type Default Description
id String auto Primary key
outputId String - FK to GeneratedOutput (1:1)
jobId String? - FK to GenerationJob (1:1)
workflowInfo Json? - {id, name, version}
modelInfo Json? - {name, version, hash}
lorasUsed Json [] [{name, version, weight}]
parameters Json {} Generation parameters
inputs Json [] [{outputId, url, type}]
seed Int? - Random seed
generatedAt DateTime now() Generation timestamp
generationDuration Int? - Duration in ms
gpu String? - GPU model
provider String? - Provider (cloud/local)
cost Float? - Generation cost

Table: provenance

ModelRegistry#

Registry of AI models, checkpoints, LoRAs, VAEs, and embeddings used for generation.

Field Type Default Description
id String auto Primary key
name String - Model name
description String? - Description
type ModelType - Model type
format ModelFormat - File format
version String - Version string
storagePath String - Storage path
sizeBytes BigInt - File size
hashAlgorithm HashAlgorithm SHA256 Hash algorithm
hashValue String - File hash
baseModel String? - Base model identifier
trainedOn String? - Training data description
triggerWords String[] [] Trigger words (for LoRA)
tags String[] [] Tags
metadata Json {} Metadata
compatibleEngines WorkflowEngine[] [] Compatible engines
usageCount Int 0 Usage count
downloadCount Int 0 Download count
starCount Int 0 Star count
uploaderId String - Uploader user ID
organizationId String? - Organization
isPublic Boolean false Public visibility
isActive Boolean true Active status
isVerified Boolean false Verification status

Unique: [name, version] Table: model_registry

RetentionPolicy#

Defines data lifecycle policies that control how long generated outputs stay in each storage tier before archival or deletion.

Field Type Default Description
id String auto Primary key
name String - Policy name (unique)
description String? - Description
hotRetentionDays Int 30 Days in hot storage
warmRetentionDays Int 90 Days in warm storage
coldRetentionDays Int 365 Days in cold storage
totalRetentionDays Int 730 Total retention days
archiveAfterInactivityDays Int? - Auto-archive after N days
outputTypes OutputFileType[] [] Applies to these types
tags String[] [] Filter by tags
isActive Boolean true Active status
organizationId String? - Organization scope

Table: retention_policies

GpuWorker#

Registration record for GPU worker nodes that process generation jobs.

Field Type Default Description
id String auto Primary key
name String - Worker name
hostname String - Hostname (unique)
gpuType String - GPU model
gpuMemoryMb Int - GPU VRAM in MB
gpuCount Int 1 Number of GPUs
supportedTypes GenerationType[] [] Supported generation types
supportedEngines WorkflowEngine[] [] Supported engines
maxConcurrentJobs Int 1 Max concurrent jobs
status String "offline" Worker status
lastHeartbeat DateTime? - Last heartbeat
currentJobId String? - Current job ID
totalJobsProcessed Int 0 Total jobs processed
totalGpuTimeSeconds BigInt 0 Total GPU seconds used
averageJobDuration Float? - Average job duration
metadata Json {} Metadata

Table: gpu_workers

Key Enums#

GenerationType#

AI generation types: TEXT_TO_IMAGE, IMAGE_TO_IMAGE, TEXT_TO_VIDEO, IMAGE_TO_VIDEO, TEXT_TO_3D, IMAGE_TO_3D, TEXT_TO_AUDIO, VOICE_SYNTHESIS, MUSIC_GENERATION, UPSCALING, INPAINTING, BLENDER_RENDER, GAUSSIAN_SPLATTING, MESH_PROCESSING, TEXTURE_UPSCALE

WorkflowEngine#

Workflow execution engines: COMFYUI, BLENDER, UNREAL, GODOT, CUSTOM

WorkflowCategory#

Workflow categories: IMAGE_GENERATION, VIDEO_GENERATION, AUDIO_GENERATION, GENERATION_3D, UPSCALING, STYLE_TRANSFER, INPAINTING, COMPOSITING, RENDERING, SIMULATION, UTILITY, OTHER

ModelType#

AI model types: CHECKPOINT, LORA, CONTROLNET, VAE, EMBEDDING, UPSCALER, CUSTOM

ModelFormat#

Model file formats: SAFETENSORS, PYTORCH, ONNX, GGUF, BLEND, GLTF, FBX, OTHER

OutputFileType#

Output categories: IMAGE, VIDEO, AUDIO, MODEL_3D, POINT_CLOUD, ANIMATION, TEXTURE, MATERIAL, DOCUMENT, DATA, OTHER

StorageTier#

Storage tiers for lifecycle management: HOT (SSD fast access), WARM (standard), COLD (archive), GLACIER (deep archive)

LineageEdgeType#

Output derivation types: DERIVED_FROM, COMPOSED_OF, REFINED_FROM, UPSCALED_FROM, CONVERTED_FROM

HashAlgorithm#

Checksum algorithms: SHA256, SHA512, MD5, BLAKE3