#orcarouter #anthropic #deepseek #gemini

OrcaRouter Provider

OrcaRouter provider plugin for OpenClaw — adaptive routing across many LLMs through a single OpenAI-compatible API.

Install

openclaw plugins install clawhub:@continuum-ai-corp/openclaw-orcarouter

@continuum-ai-corp/openclaw-orcarouter

OrcaRouter provider plugin for OpenClaw — adaptive routing across many LLMs through a single OpenAI-compatible API (https://api.orcarouter.ai/v1).

I'm an engineer on the OrcaRouter team.

Install

Requires OpenClaw 2026.5.12 or newer (Node ≥ 22).

bash

openclaw plugins install clawhub:@continuum-ai-corp/openclaw-orcarouter
# or, during launch cutover, from npm directly:
openclaw plugins install @continuum-ai-corp/openclaw-orcarouter

Configure

bash

openclaw onboard --auth-choice orcarouter-api-key
# or:
openclaw onboard --orcarouter-api-key "$ORCAROUTER_API_KEY"

OrcaRouter API keys start with sk-orca-. Get one at orcarouter.ai/console.

Config snippet (~/.openclaw/config.json5):

json5

{
  env: { ORCAROUTER_API_KEY: "sk-orca-..." },
  agents: {
    defaults: {
      model: { primary: "orcarouter/auto" },
    },
  },
}

Models

Model refs follow orcarouter/<vendor>/<model> for routed upstream models, plus the virtual router orcarouter/auto.

Model ref	Notes
`orcarouter/auto`	Adaptive router; strategy is configurable per workspace at orcarouter.ai/console
`openai/gpt-5.5`	OpenAI flagship
`anthropic/claude-opus-4.7`	Anthropic reasoning flagship (does not accept `temperature`)
`google/gemini-3-flash-preview`	Google preview
`deepseek/deepseek-v4-pro`	DeepSeek flagship

Free-form model strings are accepted — any orcarouter/<vendor>/<model> ID is forwarded as-is. See the full catalog at orcarouter.ai/models.

Routing

orcarouter/auto is a virtual router, not a model. Strategy options configured at orcarouter.ai/console/routing:

Strategy	Behavior
`cheapest`	Lowest-priced upstream that can serve the request (default).
`balanced`	Trades off price vs latency vs quality.
`quality`	Highest-quality upstream.
`adaptive`	LinUCB contextual bandit picks among candidates from request features.
`gated_adaptive`	Adaptive plus a task-difficulty score that gates between weak/strong pools.

extra_body (per-request routing override):

json

{ "extra_body": { "models": ["openai/gpt-5", "openai/gpt-4o"], "route": "fallback" } }

Attribution headers

On verified api.orcarouter.ai routes, the plugin sends:

Header	Value
`HTTP-Referer`	`https://openclaw.ai`
`X-Title`	`OpenClaw`

OrcaRouter accepts unknown headers (OpenAI-compat); attribution surfaces in the OrcaRouter console traffic view.

Caveats

orcarouter/auto defaults to the cheapest strategy, which may select upstreams without tool/function-calling support. For agentic flows that require tool calls, either pin a tool-capable model (e.g. orcarouter/openai/gpt-5) or adjust the auto router pool at orcarouter.ai/console/routing.
Reasoning models reject temperature:
- anthropic/claude-opus-4.7
- OpenAI gpt-5 family (incl. mini/nano)
- deepseek/deepseek-reasoner Pass reasoning controls via reasoning_effort (top-level) for OpenAI / Gemini / Grok / Qwen / Kimi reasoners; via thinking: {type: "enabled", budget_tokens: N} for Anthropic.

Development

bash

npm install
npm run test           # unit tests
LIVE=1 OPENCLAW_LIVE_TEST=1 ORCAROUTER_API_KEY=sk-orca-... npm run test:live

License

MIT