Backend API Reference

The combined launcher normally exposes the FastAPI backend at http://localhost:8005. The web UI uses port 8000. Source: OHMind_backend.py.

Endpoints

Method	Path	Behavior
GET	/	Service identity and initialized MCP/RAG information
GET	/health	Liveness response {"status":"healthy"}
GET	/info	Graph/assistant metadata
POST	/threads	Create a thread; optional metadata object
GET	/threads	Process-local thread registry
GET	/threads/{thread_id}	Process-local metadata, 404 if absent
POST	/threads/{thread_id}/runs	Execute a request and return result/tool events
POST	/threads/{thread_id}/runs/stream	Stream run events as server-sent events
GET	/assistants	Available assistant descriptions
GET, POST	/threads/{thread_id}/history	Actual LangGraph checkpoint snapshots
GET	/threads/{thread_id}/state	Actual latest LangGraph checkpoint snapshot

No memory/policy administration routes are implemented in this backend. Use the Python APIs for explicit integrations. Backend routes do not establish a multi-tenant authorization boundary; use this service within your configured deployment boundary.

Create a thread and run

curl --fail -X POST http://localhost:8005/threads \
  -H 'Content-Type: application/json' -d '{"metadata":{}}'

The response contains thread_id. Retain it for subsequent requests:

THREAD_ID=replace-with-returned-thread-id
curl --fail -X POST "http://localhost:8005/threads/$THREAD_ID/runs" \
  -H 'Content-Type: application/json' \
  -d '{"input":{"content":"Describe the available HEM tools."}}'

The non-streaming response includes run_id, thread_id, status, result, and tool_events. A completed agent run can contain a tool failure or an asynchronously started scientific job; check the result and job status separately.

RunRequest accepts input, messages, config, stream_mode, and multitask_strategy. The current handlers construct their own graph configuration; accepting a field does not mean all LangGraph server semantics are implemented. For simple input, use input.content. Alternatively, send messages with message type and string content (or text-content blocks); the handler extracts the last message. Missing content produces HTTP 400.

Streaming

Use the same body with /runs/stream and curl -N. Consume SSE data: frames rather than treating the response as one JSON object. The handler produces run metadata and token/tool/state/error events; clients should handle the event kinds they use and surface errors.

Checkpoints across restarts

curl --fail "http://localhost:8005/threads/$THREAD_ID/state"
curl --fail "http://localhost:8005/threads/$THREAD_ID/history"

Snapshots contain values, next, config, metadata, created_at, parent_config, and tasks. Unknown IDs with no checkpoint and no current registry entry return 404. With PostgreSQL checkpoints enabled, state/history remain queryable by saved ID across backend restarts. The separate thread registry is not durable; /threads can be empty and the metadata endpoint can return 404 for a thread whose graph checkpoints still exist.

See persistence configuration and verification tutorial.