Provider setup

Set up OpenAI

Standard config for an OpenAI account with the Responses wire.

Overview

OpenAI is the oldest and most battle-tested provider in Kition. We target the Responses API (/v1/responses) rather than legacy Chat Completions — it natively supports reasoning tokens, streamed tool calls, and multimodal input, which maps 1:1 to the Kition agent message format.

All GPT-5 / GPT-5.1 / o-series models run over this wire. Once entered, the API key is persisted to the macOS Keychain / Windows Credential Manager / Linux Secret Service — never plaintext on disk.

Generate an API key

Log in to platform.openai.com → API keys in the left rail
Click Create new secret key, name it "Kition Desktop"
Copy the string starting with sk-proj-... — you cannot see it again after leaving the page
We recommend a dedicated Project for this key with a monthly hard limit (Settings → Limits)

Add it in Kition

Open Settings → Providers → Add Provider → OpenAI. The template prefills baseURL and wire — you only paste the key and choose a default model.

{
  "name": "openai",
  "baseURL": "https://api.openai.com/v1",
  "apiKey": "sk-proj-...",
  "wire": "responses",
  "defaultModel": "gpt-5.1",
  "headers": {
    "OpenAI-Beta": "responses=v1"
  }
}

Auth header

Kition attaches Authorization: Bearer <key> to every request automatically. If you go through an enterprise OpenAI deployment or an Azure-style gateway you may need an extra api-key header — see the custom provider doc.

curl https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.1","input":"ping"}'

Recommended models

gpt-5.1 — current workhorse; balance of quality and latency, default for chat and agent
gpt-5 — previous flagship, one tier cheaper, great for bulk summarization
gpt-5.1-mini — fast tier for long-context / indexing jobs
o4 series — explicit reasoning; reserve for math and complex planning

Common errors

401 invalid_api_key — key wrong or revoked; paste it again
403 model_not_found — your OpenAI account has not unlocked the model; request it on Limits
429 rate_limit_exceeded — RPM/TPM cap hit; Kition retries with exponential backoff up to 3 times
429 insufficient_quota — billing balance, not rate; top up on Billing
400 context_length_exceeded — prompt too long; trim it or switch to a long-context model

Rate limits and billing

Tier 1 starts around 500 RPM / 200k TPM — plenty for a single desktop user. Long jobs (deep research, batch indexing) can hit the TPM ceiling; consider upgrading to Tier 2 or capping Concurrency to 2 in Settings → Agent.

Kition never prepays or stores billing data — every dollar lands directly on your OpenAI invoice, viewable in real time on the platform Usage page.

← Back to help center More in Provider setup →

Set up Anthropic

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

Set up AiApiy reverse-proxy

One key for OpenAI / Anthropic / Gemini — stable from China.