API Reference

Base URL

• Local HTTP: http://<host>:8092/v1
• P2P HTTP: exposed via LibP2P; use the p2p and service endpoints below to reach peers and services.

Health

GET /v1/health

Returns:

{ "status": "ok" }

Distributed Node Table (DNT)

GET /v1/dnt/table          # Full node table (CRDT view)
GET /v1/dnt/peers          # Currently connected peers
GET /v1/dnt/peers_status   # All known peers with connectedness
GET /v1/dnt/bootstraps     # Discoverable bootstrap multiaddrs
POST /v1/dnt/_node         # Update local node entry
DELETE /v1/dnt/_node       # Remove local node entry

Notes:

• bootstraps returns multiaddr strings like /ip4/<ip>/tcp/43905/p2p/<peerId> for reachable bootstrap nodes.

P2P forward

Forward an HTTP request directly to a specific peer over LibP2P:

GET|POST|PATCH /v1/p2p/:peerId/*path

Example:

curl -sS http://localhost:8092/v1/p2p/<PEER_ID>/v1/health

Local service forward

Forward to a local service by name (registered on this node):

GET|POST|PATCH /v1/_service/:service/*path

Example:

curl -sS http://localhost:8092/v1/_service/llm/v1/models

Global service forward (load balanced)

Forward to a provider in the network that offers the named service. Selection uses identity groups (e.g., model) parsed from the JSON body.

GET|POST|PATCH /v1/service/:service/*path

Example (OpenAI-compatible chat):

curl -sS -X POST \
  http://localhost:8092/v1/service/llm/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "gpt2",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Behavior:

• The dispatcher collects all providers that registered service=llm and whose identity groups include model=gpt2.
• One candidate is selected at random (internal load balancing).
• The request is forwarded over LibP2P HTTP to the chosen peer’s local service endpoint: /v1/_service/llm/*path.