Power Warp's agents from your own inference endpoint.

Quick answer: settings, URL, key, model ids.

Open Warp's Settings and search for "inference endpoint" to jump to the configuration. Enter the endpoint URL, https://api.apisrouter.com/v1, your APIsRouter key as the credential, and the model id or ids you want to route through it, for example claude-sonnet-4-6 and gpt-5.3-codex-spark. Once saved, those models appear in Warp's model picker alongside the built-in options. The billing behavior is the point: when you explicitly select an endpoint-routed model from the picker, Warp routes the request through your endpoint instead of consuming Warp's AI credits, and the cost lands on your gateway balance where you can see it per request. Warp's docs describe this flow as mirroring their Bring Your Own API Key setup, so it will feel familiar if you have configured BYOK before.

Endpoint URL:  https://api.apisrouter.com/v1
Credential:    sk-YOUR-APISROUTER-KEY
Model id(s):   claude-sonnet-4-6
               gpt-5.3-codex-spark

then: select the endpoint-routed model
      in Warp's model picker

How Warp uses the endpoint.

Warp is the agentic terminal: its agents draft commands, explain failures, run multi-step tasks, and edit code from the same window you work in. Warp's own documentation states the requirement plainly: the endpoint must implement the OpenAI Chat Completions API at /v1/chat/completions, and their launch post names model routers and hosted gateways as intended targets, so a multi-vendor gateway is a first-class citizen here, not a trick. Two documented constraints are worth knowing before you start. First, the endpoint must be publicly reachable: localhost, 127.0.0.1, and private network URLs are rejected at configuration time, which rules out pointing Warp directly at a local inference server but is irrelevant for a hosted gateway. Second, Warp's Auto model selection always consumes Warp credits even when a custom endpoint is configured; only explicitly picking an endpoint-routed model from the picker routes through your endpoint. If your spend is not moving to the gateway console, check which model the session actually used. Availability is plan-dependent: custom inference endpoints are offered on Free and eligible paid plans for individuals and organizations of ten or fewer people, while larger organizations need Warp's Business or Enterprise tiers. That comes straight from Warp's docs and can change on their side.

Choosing models for terminal agent work.

Terminal agents bill differently than chat: every command explanation, retry, and output summary is a request carrying session context. Since endpoint-routed usage lands in the gateway console per request and per model, a week of real usage gives you an honest number per candidate model, which beats guessing from a pricing page, including this one.

• claude-sonnet-4-6 as the daily driver: dependable command generation, strong tool use, good judgment on multi-step tasks.
• gpt-5.3-codex-spark for fast scoped work, where per-turn latency shapes how the terminal feels.
• claude-opus-4-7 for the hard sessions: debugging across services, long investigative chains, infrastructure changes you want reasoned through.
• gpt-5.5 as the frontier generalist to A/B against Claude on your own shell workflows.
• deepseek-v4-pro as the value pick for high-volume routine agent use where rate matters more than frontier polish.

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

Failure modes specific to Warp.

As always, prove the gateway half with two curl commands, the models listing and one chat completion, before debugging Warp's side. If curl passes and Warp fails, the problem is in the endpoint form or the model picker selection, nowhere else.

• Endpoint rejected at save time: Warp refuses localhost and private-network URLs by design. The endpoint must be publicly reachable; a hosted gateway URL passes this check.
• Spend still hitting Warp credits: the session is on Auto model selection, which always uses Warp credits. Explicitly pick the endpoint-routed model in the picker.
• Model not found: the id you typed into the endpoint config does not match the gateway catalog. Copy ids from the /v1/models listing byte for byte.
• 401 from the endpoint: the credential field holds a stale or truncated key. Curl the models listing with the same key to confirm outside Warp.
• Feature not visible at all: check your plan. Custom endpoints are available for individuals and small orgs on Free and eligible paid plans; larger organizations need Business or Enterprise per Warp's docs.

Who routes Warp through a gateway.

• Developers who want Claude behind Warp's agents on a prepaid balance, with a free start and no card.
• Heavy agent users whose Warp credit consumption outgrew their plan, moving explicit model picks onto their own metered endpoint.
• Model shoppers comparing coding models on real terminal workflows, where each candidate is one more id in the endpoint config.
• Teams of ten or fewer standardizing on one gateway key so terminal, editor, and CI agents share a single usage log.
• People already running other tools through a gateway who want Warp on the same key and console.

Verify the endpoint and confirm routing.

Run the two standard checks with the exact key and id you configured in Warp. Both passing means the gateway is ready and any remaining issue is in Warp's settings or model selection. Then run one real agent task in Warp with the endpoint model explicitly selected, and confirm the request appears in the APIsRouter console with the expected model and token counts. That single confirmation catches the Auto-mode gotcha immediately, and from then on the console is your per-request record of what the terminal actually spends.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | head -50

curl -s https://api.apisrouter.com/v1/chat/completions \
  -H "Authorization: Bearer $APISROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-4-6",
       "messages":[{"role":"user","content":"ping"}]}'

FAQ