Integrations

Tool integrations

Warp accepts OpenAI Chat Completions, Anthropic Messages and OpenAI Responses, so an existing client works by changing a base URL and a key. All three run the same gateway, so routing, savings, guardrails and the audit trail apply identically.

Which endpoint

Most tools use the OpenAI format. Claude Code speaks only Anthropic Messages and Codex speaks only OpenAI Responses, so each gets a dedicated endpoint — no converter to install.

Format	Endpoint	Auth header	Tools
OpenAI	`https://api.warp.inc/v1`	`Authorization: Bearer`	Cursor, OpenClaw, Hermes, AI SDK, OpenAI SDK
Anthropic Messages	`https://api.warp.inc`	`x-api-key`	Claude Code, Anthropic SDK
OpenAI Responses	`https://api.warp.inc/v1`	`Authorization: Bearer`	Codex CLI

Setup

Pick the tool you use and its own setup appears. The format, the endpoint and the config path are filled in already.

This tool speaks Anthropic Messages and nothing else → https://api.warp.inc/v1/messages

Have the agent do it

Paste this into the tool itself — it applies the change and sends a real request to confirm.

Claude Code · agent prompt

Point Claude Code at Warp as its model provider.

- Base URL (not the full request path): https://api.warp.inc
  ← Claude Code appends /v1/messages itself — do not include it
- API key: sk-warp-...  ← replace with my key
- Model: warp/code
- Config lives at: ~/.claude/settings.json

Put it in that config file rather than a shell export — an export lives only in
the tab it was typed in, which is a common way for this to silently not apply. When you're done run `claude -p "/status"` and confirm it took.

Or set it up by hand

Install

terminal

npm install -g @anthropic-ai/claude-code

Configure

~/.claude/settings.json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.warp.inc",
    "ANTHROPIC_API_KEY": "sk-warp-...",
    "ANTHROPIC_MODEL": "warp/code",
    "ANTHROPIC_SMALL_FAST_MODEL": "warp/auto"
  }
}

Create an API key →

Verify

terminal

claude
# then, inside Claude Code:
/status

You should see — Anthropic base URL: https://api.warp.inc · Model: warp/code

This tool speaks OpenAI Responses and nothing else → https://api.warp.inc/v1/responses (Codex has no other option)

Have the agent do it

Paste this into the tool itself — it applies the change and sends a real request to confirm.

Codex CLI · agent prompt

Point Codex CLI at Warp as its model provider.

- Base URL (not the full request path): https://api.warp.inc/v1
  ← Codex CLI appends /responses itself — do not include it
- API key: sk-warp-...  ← replace with my key
- Model: warp/code
- Config lives at: ~/.codex/config.toml

Put it in that config file rather than a shell export — an export lives only in
the tab it was typed in, which is a common way for this to silently not apply. When you're done run `codex exec "reply with OK"` and confirm it took.

Or set it up by hand

Install

terminal

npm install -g @openai/codex

Configure

~/.codex/config.toml

model_provider = "warp"
model = "warp/code"

[model_providers.warp]
name = "Warp"
base_url = "https://api.warp.inc/v1"
env_key = "WARP_API_KEY"
wire_api = "responses"

Create an API key →

Verify

terminal

export WARP_API_KEY="sk-warp-..."
codex

You should see — warp/code in the status line

This tool speaks the OpenAI format → https://api.warp.inc/v1/chat/completions

Have the agent do it

Paste this into the tool itself — it applies the change and sends a real request to confirm.

Cursor · agent prompt

Point Cursor at Warp as its model provider.

- Base URL (not the full request path): https://api.warp.inc/v1
  ← Cursor appends /chat/completions itself — do not include it
- API key: sk-warp-...  ← replace with my key
- Model: warp/code
- Config lives at: Settings → Models → OpenAI API Key → Override base URL

Put it in that config file rather than a shell export — an export lives only in
the tab it was typed in, which is a common way for this to silently not apply. When you're done, tell me what to check in the UI.

Or set it up by hand

Configure

Settings → Models → OpenAI API Key → Override base URL

Base URL:  https://api.warp.inc/v1
API Key:   sk-warp-...
Model:     warp/code

Create an API key →

Verify

terminal

Settings → Models → Verify

You should see — the key verifies and warp/code appears in the model list

This tool speaks the OpenAI format → https://api.warp.inc/v1/chat/completions

Have the agent do it

Paste this into the tool itself — it applies the change and sends a real request to confirm.

OpenCode · agent prompt

Point OpenCode at Warp as its model provider.

- Base URL (not the full request path): https://api.warp.inc/v1
  ← OpenCode appends /chat/completions itself — do not include it
- API key: sk-warp-...  ← replace with my key
- Model: warp/code
- Config lives at: opencode.json

Put it in that config file rather than a shell export — an export lives only in
the tab it was typed in, which is a common way for this to silently not apply. When you're done run `opencode run "reply with OK"` and confirm it took.

Or set it up by hand

Install

terminal

npm install -g opencode-ai

Configure

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "warp": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Warp",
      "options": {
        "baseURL": "https://api.warp.inc/v1",
        "apiKey": "{env:WARP_API_KEY}"
      },
      "models": {
        "warp/code": { "name": "Warp Code" },
        "warp/auto": { "name": "Warp Auto" }
      }
    }
  }
}

Create an API key →

Verify

terminal

opencode
# then, inside OpenCode:
/models

You should see — Warp listed as a provider, with warp/code selectable

This tool speaks the OpenAI format → https://api.warp.inc/v1/chat/completions

Have the agent do it

Paste this into the tool itself — it applies the change and sends a real request to confirm.

OpenClaw · agent prompt

Point OpenClaw at Warp as its model provider.

- Base URL (not the full request path): https://api.warp.inc/v1
  ← OpenClaw appends /chat/completions itself — do not include it
- API key: sk-warp-...  ← replace with my key
- Model: warp/code
- Config lives at: ~/.openclaw/openclaw.json

Put it in that config file rather than a shell export — an export lives only in
the tab it was typed in, which is a common way for this to silently not apply. When you're done, tell me what to check in the UI.

Or set it up by hand

Configure

~/.openclaw/openclaw.json

{
  "models": {
    "providers": {
      "warp": {
        "baseUrl": "https://api.warp.inc/v1",
        "apiKey": "sk-warp-...",
        "api": "openai-completions",
        "models": [
          { "id": "warp/code", "name": "Warp Code", "contextWindow": 1000000, "maxTokens": 8192 }
        ]
      }
    }
  },
  "agents": { "defaults": { "models": ["warp/warp/code"] } }
}

Create an API key →

Verify

terminal

openclaw gateway restart

You should see — warp/code · warp at the bottom of the Control UI

This tool speaks the OpenAI format → https://api.warp.inc/v1/chat/completions

Have the agent do it

Paste this into the tool itself — it applies the change and sends a real request to confirm.

Hermes · agent prompt

Point Hermes at Warp as its model provider.

- Base URL (not the full request path): https://api.warp.inc/v1
  ← Hermes appends /chat/completions itself — do not include it
- API key: sk-warp-...  ← replace with my key
- Model: warp/code
- Config lives at: terminal

Put it in that config file rather than a shell export — an export lives only in
the tab it was typed in, which is a common way for this to silently not apply. When you're done, tell me what to check in the UI.

Or set it up by hand

Configure

terminal

export OPENAI_BASE_URL="https://api.warp.inc/v1"
export OPENAI_API_KEY="sk-warp-..."

hermes model   # pick warp/code

Create an API key →

Verify

terminal

hermes

You should see — warp/code in the status bar

This tool speaks the OpenAI format → https://api.warp.inc/v1/chat/completions

Have the agent do it

Paste this into the tool itself — it applies the change and sends a real request to confirm.

Vercel AI SDK · agent prompt

Point Vercel AI SDK at Warp as its model provider.

- Base URL (not the full request path): https://api.warp.inc/v1
  ← Vercel AI SDK appends /chat/completions itself — do not include it
- API key: sk-warp-...  ← replace with my key
- Model: warp/auto
- Config lives at: app/api/chat/route.ts

Put it in that config file rather than a shell export — an export lives only in
the tab it was typed in, which is a common way for this to silently not apply. When you're done, tell me what to check in the UI.

Or set it up by hand

Configure

app/api/chat/route.ts

import { createOpenAI } from "@ai-sdk/openai";
import { generateText } from "ai";

const warp = createOpenAI({
  baseURL: "https://api.warp.inc/v1",
  apiKey: process.env.WARP_API_KEY,
});

export async function POST() {
  const { text } = await generateText({
    model: warp("warp/auto"),
    prompt: "Summarize this quarter's metrics",
  });
  return Response.json({ text });
}

Create an API key →

Verify

terminal

node --run dev

You should see — a completion, and the request on your Warp dashboard

This tool speaks the OpenAI format → https://api.warp.inc/v1/chat/completions

Have the agent do it

Paste this into the tool itself — it applies the change and sends a real request to confirm.

OpenAI SDK · agent prompt

Point OpenAI SDK at Warp as its model provider.

- Base URL (not the full request path): https://api.warp.inc/v1
  ← OpenAI SDK appends /chat/completions itself — do not include it
- API key: sk-warp-...  ← replace with my key
- Model: warp/auto
- Config lives at: client.mjs

Put it in that config file rather than a shell export — an export lives only in
the tab it was typed in, which is a common way for this to silently not apply. When you're done run `node client.mjs` and confirm it took.

Or set it up by hand

Configure

client.mjs

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.warp.inc/v1",
  apiKey: process.env.WARP_API_KEY,
});

const res = await client.chat.completions.create({
  model: "warp/auto",
  messages: [{ role: "user", content: "Hello" }],
});

console.log(res.choices[0].message.content);

Create an API key →

Verify

terminal

node client.mjs

You should see — a completion, and the request on your Warp dashboard

Not listed? Pick the tool with the same format, then change the tool name and config path in the prompt.

Routing

warp/auto and warp/code hand model choice over wholesale. The alternative is to anchor the model you already use as the quality ceiling and let only the easy requests route down — add :auto to the model name your config already had. A dial goes in the same place.

model name

anthropic/claude-sonnet-5        # always this model, no routing
anthropic/claude-sonnet-5:auto   # route; never above claude-sonnet-5
anthropic/claude-sonnet-5:balanced
anthropic/claude-sonnet-5:max-savings
warp/auto:balanced               # a router; a dial needs an anchor, so this one is ignored

Hard requests are then always answered by claude-sonnet-5, and only easy ones route down. The anchor is a ceiling, not a selection — nothing climbs above it. If a lower tier returns something unusable, it is automatically re-served once from the tier above (on by default).

Tools that cannot build a request body

A model name is something every tool lets you type, so tools that cannot build a custom request body — Claude Code, Cursor — use this exactly as everything else does. See Anchored routing.

Anthropic Messages directly

For clients implemented directly against the Messages format. Routing options are identical to the OpenAI endpoint — see Warp router.

terminal

curl https://api.warp.inc/v1/messages \
  -H "x-api-key: $WARP_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "warp/auto",
    "max_tokens": 1024,
    "messages": [{ "role": "user", "content": "Hello" }]
  }'

Unsupported fields

These two fields are ignored when present. Accepting a value that is never applied would misrepresent what the endpoint does, so it is stated here instead.

Field	Why
`thinking`	The gateway's internal response shape has no field for reasoning content, so an upstream trace cannot be carried back even when the model produces one. Thinking blocks in your history are accepted and ignored.
`cache_control`	There is no path to carry prompt-cache markers upstream, and a request may route to a model that is not Anthropic's.

top_k is forwarded. This table claimed for a while that it had no counterpart, which was not true — the gateway was dropping a knob the target model would have taken.

POST /v1/messages/count_tokens returns an estimate. It is computed before the request resolves to a model, so there is no single tokenizer to be exact against.