ChatAI

Admin ChatAI talks to an OpenAI- or Anthropic-compatible HTTP API. You supply the endpoint; ArchForge does not ship a vendor SDK or a default key.

Configuration

Set environment variables (or arch-forge.llm.*):

LLM_PROVIDER=openai          # or anthropic
LLM_BASE_URL=https://api.openai.com/v1
LLM_API_KEY=sk-...
LLM_MODEL=gpt-4o-mini
# anthropic only
LLM_ANTHROPIC_VERSION=2023-06-01

Any gateway that implements /v1/chat/completions (OpenAI) or /v1/messages (Anthropic) works.

GET /admin/chat/config returns provider, model, baseUrl, and configured — never the key.

LlmClientFactory.requireConfigured() refuses to call the model when the key is blank.

APIs

Method	Path	Notes
GET	/admin/chat/config	Status only
GET	/admin/chat/sessions	Current user sessions
POST	/admin/chat/sessions	Create session
GET	/admin/chat/sessions/{id}/messages	History
DELETE	/admin/chat/sessions/{id}	Delete session
POST	/admin/chat/sessions/{id}/messages	SSE: delta / done / error

Write methods require @SaCheckPermission("chatai:use"). Sessions are keyed by admin user id (LoginContext.getAdminUserId()), so one user cannot list another user's chats.

SSE

event: delta
data: Hello

event: done
data: [DONE]

Rate limit: @RateLimit 20 messages / minute / user on send.

Related

• UI: ArchForgeAdmin ChatAI page.
• Errors: CHAT_AI module (107xx) in ArchForgeSpec/specs/error-codes.md.