iblai-api-inference

Call ibl.ai's OpenAI-compatible chat endpoint: identical request/response shape to OpenAI's /v1/chat/completions, but served by your deployment and routed to whichever provider/model you name. Use it for raw completions, streamed tokens, or tool calls directly — no MCP server, no agent needed. To configure which model an agent runs on use /iblai-api-agent-llm; to converse with a deployed agent (RAG, memory, history) use /iblai-api-agent-chat.

Auth & conventions

• Header: Authorization: Api-Token $IBLAI_API_KEY on every request.
• Path var: {org} = $IBLAI_ORG (no username in the path).
• Model: always provider/model form, e.g. openai/gpt-5, anthropic/claude-sonnet-4. A bare name is rejected 400 invalid_request.
• Two hosts — streaming is async/ASGI-only:
  • Non-streaming → https://api.iblai.app/dm/api/ai-mentor/orgs/{org}/v1
  • Streaming (stream: true) → https://asgi.data.iblai.app/api/ai-mentor/orgs/{org}/v1