OpenRouter setup

The openrouter adapter is an OpenAI-compatible HTTP client for openrouter.ai — the cheapest pay-per-token routing target, complementary to the subscription-backed hermes-proxy. It speaks the standard /chat/completions API, so any model in OpenRouter’s catalog is reachable by id.

1. Get an API key

Create a key at https://openrouter.ai/keys. Export it (the adapter reads it from the environment — it is never stored in the manifest):

export OPENROUTER_API_KEY="sk-or-..."

Without the key, uh adapter check openrouter reports found: false with a “set OPENROUTER_API_KEY” message (this is also the CI-skip signal), and uh mission run --runtime openrouter fails fast with a clear plan error.

2. Add the adapter manifest

uh adapter add openrouter

This writes .harness/adapters/openrouter.yaml:

schema_version: uh.adapter.v0
id: openrouter
runtime: openrouter
status: active
config:
  runtime_config:
    endpoint: "https://openrouter.ai/api/v1"
    model: "openai/gpt-4o-mini"
    request_timeout_ms: 120000
    extra_headers: {}

runtime_config keys (strict — typos are rejected at load):

Key	Required	Default	Notes
`endpoint`	no	`https://openrouter.ai/api/v1`	OpenAI-compat base URL
`model`	yes	—	any OpenRouter model id, e.g. `openai/gpt-4o-mini`, `anthropic/claude-3.5-sonnet`, `qwen/qwen-2.5-72b-instruct`
`request_timeout_ms`	no	`120000`	per-request timeout
`extra_headers`	no	`{}`	extra request headers (lower-cased)
`referer`	no	—	sets `HTTP-Referer` (OpenRouter app ranking)
`title`	no	—	sets `X-Title` (OpenRouter app ranking)

3. Verify + run

uh adapter check openrouter                 # GET /models with your key
uh mission dry-run <mission> --runtime openrouter
uh mission run <mission> --runtime openrouter

Per-mission model override:

runtime_config_overrides:
  model: "anthropic/claude-3.5-sonnet"

4. Cost routing

openrouter carries cost_class: cheap, so uh mission run --auto will pick it when it’s the cheapest installed adapter that satisfies the mission’s runtime_requirements. Use uh mission run --auto --explain to see the decision matrix.

Notes

Live smoke is deferred in CI until an OPENROUTER_API_KEY repo secret is provided; the adapter’s unit tests are fully HTTP-mocked and need no key.
Enumerate available model ids: curl -s -H "Authorization: Bearer $OPENROUTER_API_KEY" https://openrouter.ai/api/v1/models | jq '.data[].id'.