Skip to main content

Connect Agents

The AntSeed VPR exposes the models selected by your routing policy through an OpenAI- and Anthropic-compatible API. An agent can connect in either of two ways:

  • Local agent — use the buyer API on the same computer at http://127.0.0.1:8377/v1.
  • Remote agent or hosted client — define an authenticated internet-accessible endpoint from the VPR's Agents view.

The public endpoint is useful for agents on another server, hosted development tools such as Cursor, CI workers, and custom applications that cannot reach your computer's localhost address.

Agent setup skills​

Use the maintained setup skill for the agent you want to connect:

For another agent, use its custom OpenAI or Anthropic provider settings with the base URL and API key described below. See Using the API for request formats and routing behavior.

Connect locally​

Use the local endpoint when the agent runs on the same computer as the VPR:

export ANTSEED_BASE_URL="http://127.0.0.1:8377/v1"
export ANTSEED_API_KEY="antseed-p2p"

The local buyer proxy does not validate an API key, but many agent SDKs require a non-empty value. The antseed-p2p value is only a placeholder.

Test model discovery:

curl "$ANTSEED_BASE_URL/models" \
  -H "Authorization: Bearer $ANTSEED_API_KEY"

Hermes​

Current Hermes releases store named custom endpoints under providers:. Add AntSeed in ~/.hermes/config.yaml:

model:
  provider: antseed
  default: antseed
  base_url: ""
  api_mode: chat_completions

providers:
  antseed:
    name: AntSeed
    api: http://127.0.0.1:8377/v1
    api_key: antseed-p2p
    transport: chat_completions
    extra_headers:
      originator: hermes
    default_model: antseed
    models:
      antseed:
        context_length: 200000
      minimax-m2.7:
        context_length: 200000
      kimi-k2.6:
        context_length: 256000

The antseed model follows the current VPR model picker. Add concrete IDs returned by GET /v1/models when you want them in Hermes' model menu. The Hermes Agent skill covers funding, systemd, auxiliary models, and routing in detail.

OpenClaw​

Add AntSeed under models.providers in ~/.openclaw/openclaw.json:

{
  "models": {
    "providers": {
      "antseed": {
        "baseUrl": "http://127.0.0.1:8377/v1",
        "apiKey": "antseed-p2p",
        "authHeader": true,
        "api": "anthropic-messages",
        "models": [
          {
            "id": "kimi-k2.6",
            "name": "Kimi K2.6 via AntSeed",
            "reasoning": false,
            "input": ["text"],
            "contextWindow": 256000,
            "maxTokens": 8192
          }
        ]
      }
    }
  }
}

Then run openclaw models set "antseed/kimi-k2.6" followed by openclaw gateway restart. authHeader: true makes OpenClaw send the API key as Authorization: Bearer <API_KEY>, which is required by the public gateway and also works locally. See the OpenClaw skill for the complete walkthrough and setup script.

Connect a remote agent​

When the agent cannot reach the VPR on localhost, first create an authenticated endpoint by following the Public HTTPS Tunnels guide. That guide is the canonical reference for ngrok and Cloudflare setup, authentication, supported routes, CLI commands, Cursor, and troubleshooting.

After the endpoint starts, replace the local placeholder values in the configuration above with the values shown by the VPR:

AgentPublic endpoint fieldAntSeed API key field
Hermesproviders.antseed.apiproviders.antseed.api_key
OpenClawmodels.providers.antseed.baseUrlmodels.providers.antseed.apiKey

Use the complete public base URL ending in /v1. For OpenClaw, keep authHeader: true so it sends the generated AntSeed key as a bearer token.

For another agent, use its custom OpenAI- or Anthropic-compatible provider settings. See Using the API for request formats and model routing.

Last updated on