Connect AI Coding Tools

Use PrimeRouter as your "private OpenAI / Anthropic API" inside any AI coding tool. This section has a dedicated page per tool, each with macOS / Linux / Windows commands.

Tools at a glance

Tool	Form	Protocol	Docs
Claude Code	CLI	Anthropic Messages	Setup →
OpenClaw	CLI / Gateway / Agent	Anthropic Compatible	Setup →
OpenCode	CLI / TUI	OpenAI Compatible	Setup →
Cursor	IDE	OpenAI Compatible	Setup →
Cline	VS Code / Cursor extension	OpenAI Compatible	Setup →
Crush	CLI / TUI	OpenAI Compatible	Setup →
Codex Setup Guide	Windows / macOS	OpenAI Responses / Chat	Setup →
Codex CLI	CLI	OpenAI Responses / Chat	Setup →
Vibecode	Web / Desktop	OpenAI Compatible	Setup →

Protocol compatibility

PrimeRouter exposes two protocols:

Protocol	Endpoint
OpenAI Compatible	https://primerouter.ai/v1
Anthropic Messages	https://primerouter.ai (tool appends /v1/messages)

Most tools use OpenAI; Claude Code uses Anthropic.

---

Common setup (once)

1. Sign up + get an API key

1. Visit primerouter.ai → sign up
2. Console → Tokens → Add token → pick a group → Save
3. Copy sk-xxx — referred to below as sk-your-primerouter-key.

2. The three parameters every tool needs

Param	Value
Base URL	https://primerouter.ai/v1
API Key	the sk-xxx from above
Model ID	any model id, e.g. claude-sonnet-4-6, claude-opus-4-7

3. Recommended Claude model IDs

Model ID	Best for
claude-sonnet-4-6	daily coding assistant, default pick
claude-opus-4-6	stable Opus for complex tasks
claude-opus-4-7	complex reasoning, long-context refactor

Full list at Available models.

---

Smoke test

After configuring any tool, verify the key + base URL with a one-liner:

macOS / Linux

curl https://primerouter.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-your-primerouter-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [{"role":"user","content":"ping"}],
    "max_tokens": 10
  }'

Windows PowerShell

curl.exe https://primerouter.ai/v1/chat/completions `
  -H "Authorization: Bearer sk-your-primerouter-key" `
  -H "Content-Type: application/json" `
  -d '{\"model\":\"claude-sonnet-4-6\",\"messages\":[{\"role\":\"user\",\"content\":\"ping\"}],\"max_tokens\":10}'

Expect a choices[0].message.content field in the response. If you get a 4xx:

Error	Cause	Fix
401 invalid_api_key	Key typo or disabled	Console → Tokens → check status
403 no_permission	Token group ≠ channel group	Edit token, add to the right group
404 model_not_found	Wrong model id or group can't access it	Check Available models
429 too_many_requests	Rate-limited	Upgrade plan / change group
500/502	Upstream transient failure	Check primerouter.ai status

---

Streaming vs non-streaming

All listed tools use streaming (SSE) by default. If a tool hangs:

• Disable streaming in tool settings to isolate the issue
• Check PrimeRouter Console → Logs to see if the request actually arrived
• If it arrived but the response is slow, upstream may be in cooldown — try another model

---

Costs and billing

• PrimeRouter bills by actual token usage; every call is visible in Console → Logs
• Claude family triggers Prompt Cache automatically (cache_ratio = 0.1, i.e. 90% discount on cache hits) — huge win for IDE/agent workflows
• For long-context tools (Cline / Codex / Claude Code), claude-sonnet-4-6 is the sweet spot

Details at Billing rules.