peers / join

Point any OpenAI-compatible client at https://api.halsupreme.com/v1 with a Bearer token. FREE path - no paid upstream keys required for the join examples.

• GET /v1/models - list HAL virtual models
• GET /v1/peers - HAL identity (+ enabled peers)
• POST /v1/chat/completions - chat
• Local stdio MCP only: python -m hal_mcp.gateway_server

mesh right now

Authenticated GET /v1/peers returns at least two entries when the node peer is healthy:

1. hal-supreme (kind: self) - public OpenAI-compatible face at api.halsupreme.com; the current MCP gateway is a separate local stdio developer surface, not a public HTTPS endpoint
2. hal-local-ollama (kind: openai_compat, privacy_class: LOCAL, billing_class: FREE) - node-local Ollama OpenAI-compat peer used as the second live mesh entry. Not a public internet endpoint; it health-checks on the HAL node only.

A scrubbed remote example (friend-ollama → peer.example.com) stays in config with enabled: false until a real friend URL passes health.

verify (no token printed)

curl -s https://api.halsupreme.com/health
curl -s https://api.halsupreme.com/v1/peers \
  -H "Authorization: Bearer $HAL_GATEWAY_TOKEN"

Expect data[] to include hal-supreme and an enabled second peer when the local Ollama peer is up. Without a Bearer token, /v1/peers returns 401.

add your own outbound peer

• Copy the friend-ollama shape in config/peers.json
• Put secrets only in env vars named by credential_env - never in git or this page
• Keep enabled: false until GET {base_url}/models (or peer health) succeeds
• Gateway reloads peers config per request; avoid touching cloudflared

Full guide: docs/HAL_PEER_JOIN.md | wiring log: docs/outreach/PEER-WIRING.md

about