Mission#

Wire DeepSeek's hosted API into OpenClaw as an OpenAI-compatible provider. Add aliases so agents and cron jobs can request DeepSeek models by short name.

Steps#

1. Edit ~/.openclaw/openclaw.json. Add a new entry under models.providers:

   "deepseek": {
     "baseUrl": "https://api.deepseek.com",
     "apiKey": { "source": "env", "provider": "default", "id": "DEEPSEEK_API_KEY" },
     "api": "openai",
     "models": [
       {
         "id": "deepseek-chat",
         "name": "DeepSeek Chat (V3)",
         "cost": { "input": 0.27, "output": 1.10, "cacheRead": 0.07, "cacheWrite": 0 },
         "contextWindow": 65536,
         "maxTokens": 8192,
         "reasoning": false,
         "input": ["text"]
       },
       {
         "id": "deepseek-reasoner",
         "name": "DeepSeek Reasoner (R1)",
         "cost": { "input": 0.55, "output": 2.19, "cacheRead": 0.14, "cacheWrite": 0 },
         "contextWindow": 65536,
         "maxTokens": 8192,
         "reasoning": true,
         "input": ["text"]
       }
     ]
   }
   

2. Add aliases under agents.defaults.models:

   "deepseek/deepseek-chat":     { "alias": "deepseek-chat" },
   "deepseek/deepseek-reasoner": { "alias": "deepseek-reasoner" }
   

3. Run openclaw config validate — must pass.
4. Restart gateway: systemctl --user restart openclaw-gateway.service and wait until it reports active.
5. Confirm provider visible: openclaw infer providers list 2>&1 | grep deepseek.

Acceptance criteria#

• Provider deepseek registered with both models
• openclaw config validate passes
• Aliases deepseek-chat and deepseek-reasoner resolve correctly
• Gateway restarted and healthy
• No regression on existing providers (Anthropic + Ollama)

Notes#

• cost units are USD per 1M tokens. Numbers above are DeepSeek's standard pricing as of late 2025 / early 2026 — verify against platform.deepseek.com/pricing before committing.
• The schema may be slightly different; consult openclaw config schema lookup models.providers first if validation complains.