feat: add OrcaRouter provider

[xilema2]

Issue

Fixes #10989

Context

Adds OrcaRouter as a selectable provider. OrcaRouter (https://www.orcarouter.ai) is an
OpenAI-compatible meta-router that routes a single API key across 80+ upstream models
(OpenAI, Anthropic, Google, DeepSeek, xAI, Qwen, etc.) and exposes an adaptive orcarouter/auto
router.

OrcaRouter's catalog and API base already ship in the models.dev catalog this repo consumes
(provider id orcarouter, api: https://api.orcarouter.ai/v1, env ORCAROUTER_API_KEY), so this
PR does not embed any model definitions. The only client-side wiring needed is attribution
headers — the same treatment the other OpenAI-compatible meta-routers (openrouter, vercel,
zenmux, nvidia) already get.

Disclosure: I'm an engineer on the OrcaRouter team.

Implementation

OrcaRouter is registered as a custom loader in kiloCustomLoaders (the Kilo-owned provider
module), mirroring the existing opencode/openrouter/zenmux entries:

• autoload: false — the provider activates once the user supplies ORCAROUTER_API_KEY
  (or auth). Provider id, env var, and base URL all come from the models.dev catalog entry;
  no catalog is embedded in this repo.
• It injects DEFAULT_HEADERS (HTTP-Referer / X-Title / User-Agent) so OrcaRouter can
  attribute traffic to Kilo Code. User-configured headers still take precedence via the normal
  provider-options merge.

No changes to shared upstream opencode files were required (the loader lives in
src/kilocode/provider/provider.ts, an annotation-exempt path).

Scope note (pre-existing, not addressed here): as with the other meta-routers, attribution
headers are injected on the env/auth activation path. Configuring a provider purely via
opencode.json provider.<id>.options.apiKey without the env var does not inject the default
headers, because the custom-loader pass runs before the config merge and autoload: false
providers are only merged once already loaded. This is existing cross-provider loader/config-merge
behavior (e.g. nvidia behaves identically) and is intentionally left out of this provider-addition
PR rather than changing provider-initialization semantics here.

Screenshots / Video

N/A — no visual UI change (provider wiring only).

How to Test

Manual/local verification

• ORCAROUTER_API_KEY=<key> bun dev then /models lists OrcaRouter models (sourced live from
  models.dev); selecting one routes through https://api.orcarouter.ai/v1 with the Kilo
  attribution headers.

Reviewer test steps

1. From packages/opencode/: bun test ./test/kilocode/orcarouter-headers.test.ts
2. Confirm both cases pass: default attribution headers are present on env activation, and a
   config-supplied header overrides while HTTP-Referer is preserved.

Blocked checks and substitute verification

• Full bun turbo typecheck includes @kilocode/kilo-jetbrains, which requires Java 21 (not
  available in my environment). Substitute: ran bun run typecheck scoped to packages/opencode/
  (the only package touched) — passes with no errors.

Evidence (all executed by the agent locally):

• bun test ./test/kilocode/orcarouter-headers.test.ts → 2 pass / 0 fail
• bun test ./test/provider/models.test.ts ./test/provider/provider.test.ts ./test/kilocode/nvidia-headers.test.ts → 80 pass / 0 fail (fixture change does not regress)
• bun run typecheck (packages/opencode) → no errors
• bun run lint → 0 errors
• bun run script/check-opencode-annotations.ts → pass

Checklist

• Issue linked above, or exception explained
• Tests/verification described
• Screenshots/video included for visual changes, or marked N/A
• Changeset considered for user-facing changes
• I personally reviewed the diff and can explain the changes, including any AI-assisted work.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

The implementation is minimal and correct. orcarouter is wired as a kiloCustomLoaders entry in the Kilo-owned src/kilocode/provider/provider.ts — exactly the same pattern as the existing opencode override — keeping zero diff against shared upstream files. The test mirrors nvidia-headers.test.ts faithfully and uses the real provider stack rather than mocks.

Files Reviewed (4 files)

• .changeset/orcarouter-provider.md — correct patch-level changeset, user-facing description
• packages/opencode/src/kilocode/provider/provider.ts — orcarouter loader added; autoload: false, DEFAULT_HEADERS injected
• packages/opencode/test/kilocode/orcarouter-headers.test.ts — two live tests covering attribution headers + config override
• packages/opencode/test/tool/fixtures/models-api.json — orcarouter entry added so provider-list tests can find the provider

Fix these issues in Kilo Cloud

---

_{Reviewed by claude-4.6-sonnet-20260217 · 1,075,134 tokens}

_{Review guidance: REVIEW.md from base branch main}

[github-actions]

To stay organized pull requests are automatically closed after 30 days of inactivity. If the pull request is still relevant please reopen it or create a fresh new one.