OpenClaw

OpenClaw is a self-hosted personal AI that runs on your machine and speaks back through the channels you already use (WhatsApp, Telegram, Slack, Discord, iMessage, and many more). It supports multiple model providers — add Synthorai as one of them and any chat model in your workspace becomes reachable via OpenClaw's agent config.

Before you start

You'll need (1) an OpenClaw install running locally and (2) a Synthorai API key from Console → API Keys.

Step 1 — Edit ~/.openclaw/openclaw.json

Add Synthorai under providers. Existing openai/anthropic entries can stay — OpenClaw will use whichever is referenced by agent.model.

{
  "providers": {
    "synthorai": {
      "type": "openai",
      "base_url": "https://synthorai.io/v1",
      "api_key": "sk-..."
    }
  },
  "agent": {
    "model": "synthorai/gpt-5.4-mini"
  }
}

ℹ

type: openai tells OpenClaw to talk the OpenAI Chat Completions protocol. Synthorai is fully OpenAI-compatible on /v1/chat/completions, so this works with any chat model in your workspace — not just OpenAI-hosted ones.

Step 2 — (Optional) also add a Claude-native provider

If you specifically want Anthropic's native message format (Claude-family models only), add a second provider that targets the Anthropic protocol. Note the base_url does not include /v1.

{
  "providers": {
    "synthorai": {
      "type": "openai",
      "base_url": "https://synthorai.io/v1",
      "api_key": "sk-..."
    },
    "synthorai-claude": {
      "type": "anthropic",
      "base_url": "https://synthorai.io",
      "api_key": "sk-..."
    }
  },
  "agent": {
    "model": "synthorai-claude/claude-sonnet-4-6"
  }
}

Step 3 — Restart OpenClaw

OpenClaw re-reads openclaw.json on start. After restart, send your first message from any channel OpenClaw is bound to (e.g. Slack, Telegram) — the agent will respond using your configured Synthorai model.

Recommended models

For a conversational assistant the cost-quality sweet spot is usually a fast mid-tier model:

Parameter	Type	Description
`gpt-5.4-mini`	`fast`	OpenAI; fast and low-cost. Good default.
`qwen3.5-flash`	`fast`	Alibaba Qwen; competitive price for Chinese + code + multilingual.
`claude-sonnet-4-6`	`balanced`	Anthropic; best for longer-context conversation. Requires the claude provider shape above.
`gemini-2.5-flash`	`fast`	Google; long context window for file-heavy workflows.

Troubleshooting

Field name mismatches across OpenClaw versions — the fields shown above (type / base_url / api_key) are the common pattern. If a specific OpenClaw release renames them (e.g. provider_type / baseUrl), consult the official config reference and swap in the same values.

401 / 403 — API key wrong, disabled, or out of quota. Check /console/keys and /console/billing.

"Unknown model" — the model slug must match the id column at /console/pricing. If you reference it via agent.model = "synthorai/MODEL_ID" make sure MODEL_ID is routable by the key you provided.

Private / air-gapped deployments — the same JSON format works; just replace the public https://synthorai.io with your internal gateway hostname.

To see what each option costs for an always-on assistant, run your expected volume through the LLM API cost calculator.