Warp custom inference
OpenAI Chat Completions, public URL
As of 2026-10-07 Warp documents custom OpenAI-compatible inference (POST /v1/chat/completions). Next: Settings → search inference endpoint, paste https://api.qcode.cc/openai/v1 and a QCode key, copy a GPT or GLM / Kimi / DeepSeek / Qwen model id from /models (Claude does not work here), then pick that model — not Auto.
Updated 2026-10-07
Point this tool at the QCode endpoint — one key for Claude, GPT and Chinese models.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard / AMEX), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Takeaways
Protocol the endpoint must implement
Official: a custom inference endpoint must implement the OpenAI Chat Completions API.
Warp AI credits when that model is selected
Official: selecting an endpoint-routed model does not consume Warp AI credits; the endpoint bills inference. Auto still consumes Warp credits.
Network rule
Official: requests go through Warp’s servers, so localhost / private URLs are rejected. QCode's https://api.qcode.cc/openai/v1 is public HTTPS.
Docs fetched
docs.warp.dev custom-inference-endpoint page first fetched 2026-08-30, re-checked 2026-10-07 (marked Last updated Oct 6, 2026); BYOK page fetched 2026-08-30.
What custom inference is
Warp is a terminal with an Agent. Official docs define a custom inference endpoint as pointing the Agent at a public URL you control that implements OpenAI Chat Completions (OpenRouter, LiteLLM, z.ai, an internal gateway). That is not the same settings widget as first-party OpenAI/Anthropic/Google BYOK. Cloud Agents cannot see locally stored custom endpoints.
What people still search
In 2026-08 Warp’s account still talks BYOK and custom inference; a Warp engineer wrote on 2026-08-13 that custom inference URLs work for the agent in the Warp terminal, with the CLI still in progress. This site had no Warp page. Here: how the official endpoint is added, and not to click Auto after you paste QCode.
Timeline
Warp’s blog announced BYOK on Free and custom inference endpoints compatible with OpenAI Chat Completions.
Settings, search inference endpoint: base URL that exposes /v1/chat/completions, credentials, model ids. Selecting that model skips Warp AI credits.
This fetch still requires a public URL; localhost is rejected. Auto always burns Warp credits. Keys live in the local keychain and pass through Warp’s backend in-flight; official text says they are not stored on Warp servers.
Confirmed vs watch-outs
Confirmed
Custom inference is on Free and eligible paid plans (individuals / orgs of 10 or fewer per Warp’s terms). Protocol: OpenAI Chat Completions. Selecting that model does not consume Warp AI credits. Docs re-checked 2026-10-07.
Watch-outs
Official: Auto still consumes Warp credits. localhost is rejected. Cloud Agents cannot use a locally stored custom endpoint and still burn Warp credits. A ChatGPT or Claude consumer subscription cannot be connected as Warp BYOK; use an API key.
Custom endpoint vs first-party BYOK
Custom inference (this page)
Any public OpenAI-compatible URL. For QCode: https://api.qcode.cc/openai/v1, for GPT and GLM / Kimi / DeepSeek / Qwen.
First-party BYOK
OpenAI / Anthropic / Google accounts only. Gateways, LiteLLM, and QCode use the custom-inference widget, not the first-party BYOK boxes.
What to do next
① In Warp, open Settings and search inference endpoint. ② Endpoint URL: https://api.qcode.cc/openai/v1 (Warp asks for the base URL that exposes /v1/chat/completions, and its own example also ends in /v1). API key: your QCode key. Model id: a GPT or GLM / Kimi / DeepSeek / Qwen id from /models (e.g. gpt-6-sol, glm-5.3). ③ Save, then pick that model in the picker — not Auto. Claude models are not available on this OpenAI-compatible endpoint; for Claude, run Claude Code in the Warp terminal with ANTHROPIC_BASE_URL set to https://api.qcode.cc/api.
On QCode
https://api.qcode.cc/openai/v1 is QCode's public OpenAI-compatible endpoint, which matches Warp’s “must be publicly reachable” rule; it serves GPT and GLM / Kimi / DeepSeek / Qwen, while Claude only runs on the Anthropic endpoint https://api.qcode.cc/api. Live model ids are /models. On Business/Enterprise, local agent runs may still consume Warp platform credits — that is Warp’s platform meter, not QCode’s price list.
FAQ
Can Warp talk to a gateway like QCode?
Yes, for GPT and GLM / Kimi / DeepSeek / Qwen. Official custom inference is for OpenAI-compatible URLs such as OpenRouter, LiteLLM, and internal gateways; for QCode use https://api.qcode.cc/openai/v1. Claude is not available on this endpoint.
Can I use localhost?
Official docs reject localhost and private URLs because requests go through Warp’s servers. Expose a public HTTPS URL first.
Why does Auto still burn Warp credits?
Official: Auto depends on Warp’s routing, so it always consumes Warp credits even with a custom endpoint configured. To use your endpoint, select that model in the picker.
Can I plug ChatGPT Plus into Warp?
Official BYOK page: a ChatGPT or Claude consumer subscription cannot be connected to Warp. Use an API key and pay the provider.
Does Warp keep my key?
Official: stored in the local keychain; sent through Warp’s backend in-flight to call your endpoint, then discarded, not stored on Warp servers.
Where do model ids come from?
The live /models catalog.
Sources
Warp Custom inference endpoint (docs.warp.dev/agents/inference/custom-inference-endpoint/, marked Last updated Oct 6, 2026), fetched 2026-08-30 and re-checked 2026-10-07; Bring Your Own API Key fetched 2026-08-30. Launch post: warp.dev/blog/bring-your-own-inference-to-warp (2026-05-20). QCode endpoint values follow the docs.qcode.cc endpoints page (fetched 2026-10-07).
Add a custom endpoint in Warp
Public URL https://api.qcode.cc/openai/v1 for GPT and GLM. Pick that model, not Auto.
Related
Claude Desktop third-party API guide
Configure a Gateway in developer mode: Anthropic Messages API only, Claude models only, no ANTHROPIC_BASE_URL, plus the Claude Code differences.
Claude Code VS Code Extension: API Key and Custom Gateway
Credentials in claudeCode.environmentVariables, settings.json shared with the CLI, plus login prompt and model selection fixes.
Codex CLI guide
OpenAI’s official CLI.
Not affiliated with Warp. Behavior follows docs.warp.dev. This page is not a guide to bypassing Warp or any model-vendor subscription auth.