CLI Adapters

Provider Discovery

OpenClaw auto-detects available providers from environment variables:

Check which providers are active:

mekong adapters/list
# ✓ OpenRouter  (primary)
# ✓ Anthropic   (fallback 1)
# ✓ Ollama      (offline fallback)
# ✗ DeepSeek    (no API key)

Routing Logic

Tasks are routed based on three variables:

export LLM_BASE_URL=https://openrouter.ai/api/v1
export LLM_API_KEY=sk-or-v1-yourkey
export LLM_MODEL=anthropic/claude-sonnet-4

Override per-command with --model:

mekong cook "Write pitch deck" --model google/gemini-2.0-flash
mekong code "Refactor auth module" --model deepseek/deepseek-r1

Configure routing rules in mekong/adapters/llm-providers.yaml:

routing:
  coding: anthropic/claude-sonnet-4
  writing: google/gemini-2.0-flash
  analysis: deepseek/deepseek-r1
  default: openai/gpt-4o

Automatic Failover

If the primary provider fails (rate limit, outage, timeout), OpenClaw automatically falls back:

Fallback chain:
OPENROUTER_API_KEY → DASHSCOPE_API_KEY → DEEPSEEK_API_KEY
→ ANTHROPIC_API_KEY → OPENAI_API_KEY → GOOGLE_API_KEY
→ OLLAMA_BASE_URL → OfflineProvider

Configure failover behavior:

# mekong/adapters/failover.yaml
strategy: sequential   # sequential | random | latency-based
timeout_ms: 10000
max_retries: 2
notify_on_failover: true

The OfflineProvider runs a minimal Ollama model locally — ensuring tasks never fully fail even without internet.

Swarm Mode

Run multiple providers in parallel and pick the best result:

mekong cook "Generate 5 headline variants" --swarm

Swarm mode spawns the task across all available providers simultaneously, then uses a judge model to select the best output. Consumes MCU for each provider used.

Configure swarm strategy:

# mekong/adapters/swarm.yaml
mode: best-of-n        # best-of-n | consensus | fastest
n: 3                   # number of providers
judge_model: anthropic/claude-sonnet-4

Custom Adapters

Add any OpenAI-compatible API as a custom adapter:

mekong adapters/add \
  --name "my-llm" \
  --base-url https://api.my-llm.com/v1 \
  --api-key $MY_LLM_KEY \
  --model my-model-v1

Or create a config file at mekong/adapters/custom/my-llm.yaml:

name: my-llm
base_url: https://api.my-llm.com/v1
api_key_env: MY_LLM_KEY
model: my-model-v1
context_window: 128000
supports_tools: true

Custom adapters participate in failover and swarm routing automatically.