Provider setup

Set up Anthropic

Official Anthropic API + Messages wire — drive the agent with Claude.

Why Anthropic

The Claude family wins on long context, tool-call accuracy, and writing quality in our internal evals. We use the official Messages API (/v1/messages), wire id anthropic_messages. It is not the same shape as OpenAI Responses — Kition Agent runs an internal adapter so the UI looks identical.

Anthropic also has the most mature prompt-caching story right now: a cache hit saves 90% of input cost, and long system prompts get cache_control markers applied automatically.

Generate an API key

console.anthropic.com → Settings → API Keys
Create Key, name "Kition Desktop", pick a Workspace
Copy the sk-ant-api03-... string
Workspaces support spend caps to prevent runaways

Add it in Kition

Settings → Providers → Add → Anthropic. The template fills wire and the anthropic-version header for you.

{
  "name": "anthropic",
  "baseURL": "https://api.anthropic.com",
  "apiKey": "sk-ant-api03-...",
  "wire": "anthropic_messages",
  "defaultModel": "claude-sonnet-4-6",
  "headers": {
    "anthropic-version": "2023-06-01",
    "anthropic-beta": "prompt-caching-2024-07-31"
  }
}

Auth header

Anthropic uses x-api-key (not a Bearer token) and requires the anthropic-version header — Kition injects both automatically. A 401 invalid x-api-key usually means stray whitespace or that you pasted the key into the wrong field.

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "messages": [{"role":"user","content":"ping"}]
  }'

Pick a model

claude-sonnet-4-6 — default daily driver for writing and standard agent runs
claude-opus-4-7 — heavy lifting (deep reasoning, complex refactors, research synthesis); 5x the cost
claude-haiku-4-5 — fast tier for summarization, classification, lightweight subagents
All models ship 200k context; Pro users can enable 1M context beta via anthropic-beta: context-1m-2025-08-07

Common errors

401 authentication_error — key wrong or has whitespace
403 permission_error — workspace lacks the model or region is restricted
404 not_found_error — model id typo (note: claude-sonnet-4-6, not claude-4.6-sonnet)
429 rate_limit_error — RPM/ITPM/OTPM hit; Kition honors retry-after
529 overloaded_error — Anthropic-side congestion; auto-retried 3 times before surfacing
400 invalid_request_error — usually first message is not user, or tool_use/tool_result pairs are unbalanced

Rate limits and billing

Anthropic rate-limits on three axes: RPM (requests/min), ITPM (input tokens/min), OTPM (output tokens/min). Tier 1 lands around 50 RPM / 50k ITPM / 10k OTPM — fine for one user but ITPM hits first on batch indexing.

Strongly recommend prompt caching: long system prompts cost 90% less on cache hit, which dominates agent loops that resend the system block every turn. Kition enables it by default — no manual config needed.

← Back to help center More in Provider setup →

Set up OpenAI

Standard config for an OpenAI account with the Responses wire.