configure-api

name	configure-api
description	Configure OpenClaw model API providers with minimal user input. Use when user wants to add, switch, or connect LLM APIs (e.g., "接入百炼", "配置DashScope", "用DeepSeek", "换个模型"). Auto-detects provider settings from known providers database — user only needs to provide API key and optionally which plan/model they want.

Configure API

One-step LLM API provider setup for OpenClaw. User provides minimal info, you handle the rest.

Workflow

1. Identify provider — from user's request, match against known providers in {baseDir}/references/providers.json
2. Ask only what's missing — typically just the API key. If the provider has multiple plans (e.g., DashScope regular vs Coding Plan), ask which one.
3. Test → Apply → Verify — all in one go, no back-and-forth.

Step 1: Identify Provider

Read {baseDir}/references/providers.json for the full provider database. Match the user's intent to a provider entry. Users may say things like:

• "接入百炼" / "用阿里百炼" → dashscope
• "用 DashScope Coding Plan" → dashscope-coding
• "配置 DeepSeek" → deepseek
• "接入 MiniMax" / "海螺" → minimax
• "用硅基流动" / "SiliconFlow" → siliconflow
• "用 OpenAI" → openai
• "用火山引擎" / "豆包" → volcengine

If the provider isn't in the database, fall back to manual: ask for baseUrl, model ID, and API key.

Step 2: Ask Only What's Missing

From the matched provider entry you already know: baseUrl, api type, default model, context window, etc.

Typically only ask:

• API key (always required)
• Plan variant (only if provider has multiple, e.g., DashScope regular vs coding)
• Model choice (only if provider offers multiple models AND user didn't specify)

Do NOT ask for baseUrl, api type, maxTokens, contextWindow, etc. — these come from the provider database.

Step 3: Test → Apply → Verify

Run all three in sequence. Stop and report if any step fails.

# 1. Test connection
python3 {baseDir}/scripts/configure_api.py test \
  --base-url "BASE_URL" \
  --api-key "API_KEY" \
  --model "MODEL_ID" \
  --api-type "API_TYPE"

# 2. Apply config (only if test passed)
python3 {baseDir}/scripts/configure_api.py apply \
  --base-url "BASE_URL" \
  --api-key "API_KEY" \
  --model "MODEL_ID" \
  --api-type "API_TYPE" \
  --provider "PROVIDER_NAME" \
  --set-default \
  --alias "ALIAS" \
  --context-window CONTEXT_WINDOW \
  --max-tokens MAX_TOKENS

# 3. Verify hot reload (~10s after apply)
grep "gateway/reload" /tmp/openclaw/openclaw-$(date +%Y-%m-%d).log | tail -3

Success: config hot reload applied in logs. Failure: config reload skipped (invalid config) — read error, fix, retry.

If hot reload fails due to unrecognized config keys, run openclaw doctor --fix first.

Troubleshooting

• Valid api values: openai-completions, openai-responses, anthropic-messages, google-generative-ai, ollama. Never use openai-chat.
• Different API keys per plan: Some providers (e.g., DashScope) use different keys for different plans. Coding Plan keys start with sk-sp-.
• Model self-identifies wrong provider: This is normal model behavior, not a config issue. Check logs for provider= to confirm routing.