Sign in

Issues and alerts

AI fix

The AI fix panel on an issue page sends the latest event's stack trace, breadcrumbs and request (query string removed) to Claude and returns a summary, the likely cause, a fix and a confidence level. It then writes a copy-ready prompt for Claude Code or Cursor. Error text is framed as untrusted data so it cannot steer the model.

1Buy credit

On Usage, buy an AI credit pack of $5, $20 or $50 through a one-time Stripe Checkout. The balance is shared by the workspace and never expires.

2Run a fix

Owners and admins pick a model and press run; results appear for everyone on the issue. Claude Haiku 4.5 is fastest and cheapest, Sonnet 5 balanced, Opus 5 most thorough. It usually takes 10 to 60 seconds. One fix runs per issue at a time and three per workspace.

3How you are charged

Each fix costs 3x the model's real token cost. Before the call, the worst-case price is held from your balance; afterwards the real price is charged and the rest returned at once. A refusal, a cut-off answer or an API error releases the whole hold, and any hold older than 15 minutes is released on the next visit. Every purchase and charge is in the ledger on Usage.

Good to know

  • Typical prices, from the models' list prices times three: about $0.03 with Haiku 4.5, $0.06 with Sonnet 5 and $0.15 with Opus 5 for a 6,000-token report. Long stack traces cost more; the report is capped at 40,000 characters.
  • Nothing is sent to Anthropic unless someone runs a fix. The privacy policy names Anthropic as a processor for this.
  • Your coding agent can fetch the latest finished analysis through the MCP tool get_ai_fix.

Stuck? Email support@blipit.io. Keys and the exact DSN for each project are under API keys in app.blipit.io.