Set up AI summaries in Session Recorder

Turn recorded evidence into an issue draft when you select File GitHub issue. This article covers your own provider key; to use Proofline credits instead, see Connect Proofline.

Connect a provider#

1. Open Settings → AI summaries and choose OpenAI, Anthropic, Google Gemini or an OpenAI-compatible endpoint.
2. Use the provider key link, then paste your API key. Each provider’s settings are stored separately.
3. Select a suggested model or type an exact model ID available to your account.
4. Select Save summary settings.

Current model suggestions#

Checked October 24, 2026. Suggestions stay editable, so you can use another model available to your account.

• OpenAI: gpt-6.1-sol. Model documentation.
• Anthropic: claude-sonnet-5-5, claude-opus-5-5, claude-haiku-5-5 or claude-fable-5-1. Model catalog.
• Google Gemini: gemini-4-argon, gemini-3.8-flash or gemini-3.5-flash-lite. Model catalog.

New configurations start with GPT-6.1 Sol, Claude Sonnet 5.5 or Gemini 4 Argon. Your saved model choice is kept. Provider availability and charges depend on your account.

OpenAI-compatible servers#

Enter a base URL that supports Chat Completions, including /v1 if your server requires it, and the server’s model ID. Use HTTPS; HTTP is supported for localhost. A local server may leave the API key blank. No model is assumed for custom servers.

What is sent#

The extension sends up to 60,000 characters of text evidence directly to your selected provider. Annotations lead, followed by notes, narration and interactions. Repeated diagnostics are condensed and used only when relevant to the reported concern. Screenshots, attached file bytes and audio and video bytes are excluded. Transcript text may be included. To inspect the input, open Report options → Review text sent to the model.

When AI is configured, stopping a recording also requests a short session title from the same evidence. It appears in the Sessions list. If naming fails, the original page-based title remains, and a manual rename is kept.

Keys stay in local extension storage and are excluded from reports. Provider charges may apply. Without AI configured, you write the draft yourself. See AI issue summaries for how summaries behave.

Vocabulary for spoken annotations#

If transcription is also configured with Deepgram or ElevenLabs, selecting an annotation sends a bounded excerpt of visible headings, navigation and interface labels to your model, which returns unusual terms. The extension sends that short dictionary to the transcription provider to improve recognition of product names. Input values, editable fields and hidden content are excluded, and neither the excerpt nor the dictionary is saved. Your provider bills the extra model call, and ElevenLabs charges for keyterm prompting. If extraction fails, ordinary transcription continues.

Drafts and errors#

Filing keeps an edited or previously generated draft. Use Regenerate summary to replace it. If generation fails, check the key, exact model ID, endpoint and quota, or write the summary yourself. Transcription has its own provider settings.