- Needs
- Your own model from OpenAI, Anthropic, Google or an OpenAI-compatible server, or a Proofline connection
- Sends
- Up to 60,000 characters of text evidence. No screenshots, files, audio or video.
- Help
- Set up AI summaries
What it is for
Most of the time spent filing a bug goes into turning evidence into prose. The summary step does that from the recording and keeps your intent in charge: annotations and notes come first, then narration, then interactions. Repeated console and network diagnostics are condensed and kept separate as supporting evidence, so an issue about a wrong total does not turn into an issue about an unrelated warning.
Supported models
- OpenAI, Anthropic and Google Gemini through their native APIs.
- OpenAI-compatible servers that implement Chat Completions, including local ones. HTTPS is required except for
localhost, and a local server can leave the key blank. - Proofline (preview), which picks a model by default and bills each request in Proofline credits. Advanced settings let you choose the model yourself.
With your own key, you enter the model ID available in your account. Each provider keeps its own configuration, so switching does not lose settings.
How it behaves
- File GitHub issue generates the summary automatically when a model is configured. There is no separate review step, and Review text sent to the model shows the exact input.
- Drafts you have edited, and drafts generated earlier, are kept. Regenerate summary replaces them only when you ask.
- When you stop a recording, the same bounded evidence produces a short session title. The report opens immediately and the name appears when ready. A failed request keeps the original title, and a late reply never overwrites a name you typed.
- While annotating with transcription on, the same model extracts unusual terms from visible page labels to help speech recognition. See page vocabulary.
- Without a model, you write the issue yourself; everything else works the same.
Data and costs
With your own key, evidence goes directly from your browser to the provider you configured, not through Proofline. The provider’s pricing, rate limits and retention policies apply. With a Proofline connection, evidence goes to Proofline and each request uses credits; when credits run out, AI pauses until you add credits or switch to your own key. Generated text can be wrong: read the summary before you rely on it, especially in a public repository.
Record your next bug instead of describing it
Session Recorder is free and open source under the AGPL-3.0. Recording and ZIP export need no account. AI, transcription and publishing are optional: use your own providers, whose charges apply, or use credits with the optional Proofline connection preview when it reaches your installed version.