Use your own API key
Connect your own AI provider to lift the monthly limit — you pay your provider directly.
If ten AI investigations a month is not enough, connect your own provider key. Requests then go to your account, are not counted against your plan allowance, and are billed to you by that provider.
Supported providers
| Provider | Key looks like | Suggested models |
|---|---|---|
| OpenAI | sk-… | gpt-4o-mini, gpt-4o, o4-mini |
| Anthropic | sk-ant-… | claude-sonnet-4-6, claude-opus-4-8, claude-haiku-4-5 |
| Google Gemini | AIza… | gemini-2.5-flash, gemini-2.5-pro |
| OpenRouter | sk-or-v1-… | openai/gpt-4o-mini, anthropic/claude-sonnet-4-6 |
| NVIDIA | nvapi-… | openai/gpt-oss-20b, meta/llama-3.1-8b-instruct |
The model field is free text, so any model id your provider accepts will work.
Check the model id is still current
The suggestions above are the ones the app offers, and providers retire models faster than the
list is updated — NVIDIA already returns 410 Gone for some of them. If an investigation fails
immediately, a dead model id is the first thing to check: use Test Connection after changing
it.
Connecting one
Open Settings → AI Provider
Two cards: ChangeTrace AI and Use your own AI provider. Pick the second.
Choose a provider and paste the key
Optionally set a model. Leave it blank for the provider's default.
Test Connection
This makes a real call to your provider. Fix any error before saving — a wrong key saved now becomes a failed explanation later.
Save
The key is encrypted at rest and never shown again — you will see •••••••• and the last four
characters.
What changes once it is on
| ChangeTrace AI | Your own key | |
|---|---|---|
| Counts against your plan allowance | Yes | No |
| Who pays | Included | You, via your provider |
| Key visible after saving | — | No, only last 4 |
| Scope | Your organization's plan | Just you |
Your key is yours alone
It is stored against your user, not your organization. Teammates keep using ChangeTrace AI and their own allowance unless they add their own key. Nobody else can see or use yours.
Your provider will bill you
Your API provider may charge you for AI usage. ChangeTrace does not control your provider's pricing or usage limits.
Set a spending cap with your provider if that matters to you. ChangeTrace cannot see your balance and cannot stop a charge.
Limits that still apply
Unmetered is not unlimited. A technical backstop of roughly 200 requests a day and 2,000 a month per user stops a runaway script. Normal use will never approach it.
Hit it and the message is deliberately different from the plan-limit one — it does not suggest adding a key, because you already have.
Removing it
Remove API Key in the same card. You go straight back to ChangeTrace AI and your plan allowance, and the stored key is deleted.
Choosing a model
- Cheap and fast — gpt-4o-mini, gemini-2.5-flash, claude-haiku
- Better reasoning on messy incidents — gpt-4o, claude-sonnet, gemini-2.5-pro
Incident explanations are short, structured tasks. A small model usually does fine; reach for a larger one when incidents involve many overlapping changes.
{ }For developers
Endpoints are GET/PUT/DELETE /api/v1/account/ai-provider plus
POST /account/ai-provider/test, all Clerk-guarded and scoped to the authenticated user id — a
user can only read or write their own. The key is encrypted at rest and never returned by the
API. Requests with a BYOK key are logged for usage visibility but bypass plan quota accounting.