Fork of Agent Desktop by BaLaurent — adds extended thinking capture, display, and export.
Claude's extended thinking produces a reasoning trace before every response. Stock Agent Desktop discards it. This fork captures it, stores it in the local database, and gives you a button to read and export it.
- 💭 Thinking capture — Claude's chain-of-thought reasoning is captured from the SDK and persisted in the SQLite database
- In-app thinking display — collapsible thinking section on every assistant message, with word/char count
- Copy and export — copy thinking to clipboard or save as
.txtwith one click - Conversation export — markdown and JSON exports include thinking blocks
- API key login — paste an Anthropic API key directly on the login screen (no
claude loginrequired) - Separate identity — different app ID, name, and config directory so it runs alongside stock Agent Desktop
Everything else — chat, tools, MCP, file explorer, git panel, scheduler, themes — works exactly as upstream.
- Get an API key from https://console.anthropic.com/
- Launch the app and paste the key on the login screen
- Go to Settings → AI → set Max Thinking Tokens to
10000 - Start a conversation — you'll see 💭 Thinking above each response
git clone https://github.com/DocterJac/agent-desktop-thinking.git
cd agent-desktop-thinking
npm install
npm run build
npm run dist:macInstall:
cp -R release/mac-arm64/Agent\ Desktop\ \(Thinking\).app /Applications/
xattr -cr "/Applications/Agent Desktop (Thinking).app"See README-THINKING.md for full technical documentation:
- Every file changed and why
- How the SDK delivers thinking (key discovery)
- Capture and display flow diagrams
- Database schema and direct SQL queries
- Settings, rebuild instructions, known limitations
Based on Agent Desktop v0.15.0 by BaLaurent. All upstream features are preserved. See the upstream wiki for Agent Desktop's own documentation.
AGPL-3.0 — same as upstream.