A single registry of the platform's AI resources:
providers and their models, the tool catalog, and the neural networks'
system prompt — plus their management: assigning models to system
functions, chat and agent model lists, the spend aggregate. The ORM
models live in the shared core (core), Admin gives them CRUD
and screens, and the consumers — Harvester, Knowledge Store, Query
Engine, Agent Engine — read the assignments from here, with no cycles
between modules.
Architecture
Connections
ai_providers
key at-rest · connection status
providers supply models ──▶
Model catalog
ai_models
gpt-4ochat
gpt-4o-minichat
claudechat
o-serieschat
BGE-M3embed
Qwen3-0.6Bembed
chat
embedding
Assignments
model_assignments
embedding
BGE-M3
RAG
gpt-4o
Lists
chat · agent
chat_models
gpt-4o
claude
agent_models
claude
o-series
tools
tool catalog · open type registry
chat · available in chat
agents · available to agents
prompt_settings
system prompt · layered composition
safety
org
eng
runtime
API
- GET · POST/api/v1/admin/ai/providers
- GET · PATCH · DELETE/api/v1/admin/ai/providers/{id}
- GET · POST/api/v1/admin/ai/models
- PATCH · DELETE/api/v1/admin/ai/models/{id}
- GET/api/v1/admin/ai/providers/{id}/discovery
- POST/api/v1/admin/ai/providers/{id}/check
- GET · PATCH/api/v1/admin/ai/assignments
- GET · POST/api/v1/admin/ai/tools
- PATCH · DELETE/api/v1/admin/ai/tools/{id}
- POST/api/v1/admin/ai/tools/{id}/check
Detailed design decisions for the module.
The module has no frontend of its own: the registry is managed by Admin
Panel screens —
→ AI models
(providers · models · assignments),
→ AI usage
(spend by person and model, limits),
→ AI prompt
(system prompt) and
→ AI tools
(catalog and enablement).