Can I use Kilo Gateway with Claude Code?
Not officially: Kilo documents its gateway as an OpenAI-compatible chat completions API and publishes no Claude Code setup, though it answered an Anthropic-format test request on October 13, 2026.
# Not documented or supported by Kilo. It answered one test request
# in Anthropic's format on Oct 13, 2026; we did not run a full session.
export ANTHROPIC_BASE_URL=https://api.kilo.ai/api/gateway
export ANTHROPIC_AUTH_TOKEN=YOUR_KILO_API_KEY
export ANTHROPIC_API_KEY=""
export ANTHROPIC_MODEL=kilo-auto/efficient
claude
Shell exports last for one terminal; put the same values in the env block of ~/.claude/settings.json to keep them. Run /status in Claude Code to check the base URL.
What Kilo documents
| Endpoint | Documented | Notes |
|---|---|---|
| /chat/completions | Yes | OpenAI-compatible, the gateway's main API |
| /api/fim/completions | Yes | Fill-in-the-middle, Mistral models only |
| /models and /providers | Yes | Public, no key needed |
| /v1/messages (Anthropic) | No | Answered our test request in Anthropic's format |
| /v1/responses (OpenAI Responses) | No | Answered our test request in Responses format |
From kilo.ai/docs/gateway and its API reference, checked October 13, 2026. Our tests were one anonymous request each on the free auto model, with a 5-token limit; we did not run a Claude Code session.
Gotchas
- Your key is a token from Your Profile in the Kilo app, tied to your personal account.
- Kilo's auto models choose for you: kilo-auto/efficient "classifies the difficulty of each request" and routes to the cheapest model proven accurate enough; kilo-auto/frontier uses Claude; kilo-auto/free uses free models and "may route your requests to providers that log prompts and outputs".
- Because the Messages path is undocumented, tool use, prompt caching, thinking and Claude Code's newer features are untested there and could change without notice.
- Individual accounts stop when the balance reaches zero. Kilo documents spend alerts, not a cap per key.
- EU data residency is listed for Enterprise only.
What a request costs through Kilo Gateway
The catch
An endpoint that answers one test request is not a supported integration. If Kilo Code is your agent, use Kilo Gateway where Kilo documents it; if Claude Code is your agent, pick a gateway that documents Claude Code, or expect to debug it yourself.
Where Standard Compute fits
Standard Compute documents Claude Code on /v1/messages and Codex on /v1/responses. Keep Claude Code's own model names and smart routing picks a model per request, sending routine steps to efficient models; a flat plan from $19 includes the compute, stops at its budget and is the same every month. Teams and agencies get a plan per key and Europe or US rules without an Enterprise contract.
FAQ
What is the Kilo Gateway base URL?
https://api.kilo.ai/api/gateway, documented as a drop-in replacement for OpenAI's /chat/completions endpoint.
Does Kilo Gateway support the Anthropic Messages API?
Kilo's docs don't say so. A test request we sent on October 13, 2026 to /api/gateway/v1/messages came back in Anthropic's format, so it exists, but it is undocumented.
Can I use Kilo Pass credits in Claude Code?
Kilo Pass credits pay for Kilo Gateway usage, so a working gateway connection would draw on them. Kilo documents no Claude Code setup, so that use is unsupported.
Does Kilo Gateway work with Codex CLI?
Not according to Kilo's docs, which cover the reverse: signing in with a ChatGPT or Codex subscription inside Kilo. The gateway answered a Responses-format test request, but that path is undocumented.