AI Debrief

Understand AI-written code while you build.

AI Debrief explains your current git diff inside VS Code-compatible IDEs so you can keep learning what changed, why it matters, and what risks AI-assisted code may have introduced.

Supported IDEs

Works anywhere the VS Code extension host works:

• VS Code
• Cursor
• Windsurf
• VSCodium
• Other VS Code-compatible desktop IDEs

This is a desktop extension. It is not a JetBrains plugin and not a browser/mobile extension.

Fastest Setup: Local Ollama

1. Install Ollama.
2. Pull a fast model:

ollama pull phi3:mini

3. Pull a deeper model:

ollama pull deepseek-coder:6.7b

4. In your IDE, install or launch AI Debrief.
5. If local Ollama is already available, AI Debrief will try to configure it automatically on first run.
6. If you already have multiple Ollama models, AI Debrief: Setup Wizard lets you pick any installed model for fast and deep reviews.
7. Run AI Debrief: Test Connection.

What It Needs

• A folder opened as a workspace
• A git repository
• Uncommitted changes in the current file
• One configured inference backend

Backends

Ollama

• Best default for privacy and lowest setup friction
• AI Debrief tries to auto-start local Ollama when needed
• Uses vibecheck.fastModel for auto reviews
• Uses vibecheck.deepModel for manual reviews

OpenAI-Compatible

Use this for:

• OpenAI
• Groq
• OpenRouter
• Together
• Other OpenAI-style APIs

Run AI Debrief: Setup Wizard, choose OpenAI or Groq, then enter:

• base URL
• API key
• fast model
• deep model

Anthropic

Run AI Debrief: Setup Wizard, choose Anthropic, then enter:

• API key
• fast model
• deep model

Gemini

Run AI Debrief: Setup Wizard, choose Gemini, then enter:

• API key
• fast model
• deep model

Custom HTTP

Run AI Debrief: Setup Wizard, choose Custom HTTP, then enter:

• endpoint URL
• fast model label
• deep model label

Optional advanced settings:

• vibecheck.customHttpHeadersJson
• vibecheck.customHttpBodyTemplate
• vibecheck.customHttpResponseField

Secrets

Hosted-provider API keys entered through the setup wizard are stored in the IDE secret store.

Legacy plain-text settings are still read for backward compatibility:

• vibecheck.openaiApiKey
• vibecheck.anthropicApiKey
• vibecheck.geminiApiKey

Main Commands

• AI Debrief: Setup Wizard
• AI Debrief: Test Connection
• AI Debrief: Review Latest Diff (Panel)
• AI Debrief: Explain This Change
• AI Debrief: Refresh Provider Status

Main Settings

• vibecheck.reviewTrigger: auto or manual
• vibecheck.reviewScope: selection, hunk, or file
• vibecheck.provider
• vibecheck.fastModel
• vibecheck.deepModel
• vibecheck.fastTimeoutMs
• vibecheck.deepTimeoutMs
• vibecheck.fastVerbosity
• vibecheck.deepVerbosity
• vibecheck.onlyStaged

Development

npm install
npm run compile

Then press F5 in VS Code and run the extension development host.

Alpha Release Notes

• This extension is currently published as a preview alpha.
• The easiest path for testers is local Ollama.
• If Ollama is already installed and usable at http://localhost:11434, AI Debrief will try to configure itself automatically on first run.
• If local Ollama is missing or has no models, the extension guides the user toward the next step instead of failing silently.