Autonomous agents

Long-running assistant agents that route every model call through a gateway process. Both read the SayGM key from the environment:

export GM_API_KEY="gm_live_…"

Before starting a long-running agent, verify that the key has spendable credit:

curl -fsS "https://api.saygm.com/v1/credits" \
  -H "Authorization: Bearer $GM_API_KEY"

This returns the key’s current spendable balance. See check balance and current spend for the response fields. A 401 means the key is missing or invalid; a 402 from an inference call means the account cannot cover that request.

Hermes Agent

Add SayGM as a named provider in ~/.hermes/config.yaml:

providers:
  saygm:
    api: https://api.saygm.com/v1
    key_env: GM_API_KEY
    transport: chat_completions
    default_model: gpt-5.5

model:
  default: gpt-5.5
  provider: custom:saygm

Hermes reads the key from the environment variable named by key_env, so the export GM_API_KEY above covers it. To persist the key instead, put GM_API_KEY=gm_live_… in ~/.hermes/.env — Hermes keeps secrets there, never in config.yaml.

For a Claude model, use the Anthropic-compatible endpoint:

providers:
  saygm-claude:
    api: https://api.saygm.com
    key_env: GM_API_KEY
    transport: anthropic_messages
    default_model: claude-opus-4-8

Run hermes model to switch models interactively, or use the triple syntax mid-session: /model custom:saygm:gpt-5.5.

OpenClaw

Add SayGM as a custom provider in ~/.openclaw/openclaw.json:

{
  agents: {
    defaults: {
      model: { primary: "saygm/gpt-5.5" },
    },
  },
  models: {
    mode: "merge",
    providers: {
      saygm: {
        baseUrl: "https://api.saygm.com/v1",
        apiKey: "${GM_API_KEY}",
        api: "openai-completions",
        models: [
          {
            id: "gpt-5.5",
            name: "GPT-5.5 via SayGM",
            contextWindow: 272000,
            maxTokens: 128000,
          },
        ],
      },
    },
  },
}

OpenClaw expands ${GM_API_KEY} from the environment. For a Claude model, set api: "anthropic-messages", baseUrl: "https://api.saygm.com", and an available messages model ID. openclaw onboard walks through the same setup interactively — pick Custom provider when it asks for a model provider.

Check the selected model’s actual context and output limits in the model catalog rather than copying the example contextWindow and maxTokens values.