Sessions & Chat
Browse your AI coding sessions, view full conversation history with tool calls, and filter by project, source tool, or session character.
Session list
The Sessions page uses a three-panel layout: project navigation on the left, session list in the middle, and session detail on the right.
Each session row shows:
- Session title (auto-generated from conversation content)
- Source tool badge (Claude Code, Cursor, Codex CLI, Copilot CLI, VS Code Copilot Chat)
- Date and duration
- Message count, cost, and insight counts
- Session character badge (classification)
- Prompt quality score (when analyzed)
Session characters
Each session is classified into one of 7 types. Classification happens in two ways:
- During sync — a heuristic based on tool call patterns provides an initial classification
- During LLM analysis — the LLM classifies based on conversation content, providing more accurate results
The LLM classification overrides the heuristic when analysis is run.
| Character | Description |
|---|---|
deep_focus | Long, concentrated work on a specific problem or area |
bug_hunt | Debugging-driven — investigating errors, tracing issues, fixing bugs |
feature_build | Building new functionality — creating files, adding endpoints, wiring components |
exploration | Research-oriented — reading code, searching, understanding before acting |
refactor | Restructuring existing code — renaming, moving, reorganizing without new features |
learning | Knowledge-seeking — asking questions, understanding concepts, getting explanations |
quick_task | Short and focused — small fix, config change, or one-off task |
Session detail
Click any session to open the detail view with three tabs:
Insights tab
- Vitals strip — duration (with time range), messages (user/assistant split), tokens (in/cache/out breakdown), cost (with model name)
- Summary — AI-generated narrative with outcome status and key bullets (requires analysis)
- Pull Requests — GitHub PR links referenced in the conversation, shown as clickable badges
- Learnings — collapsible items with root cause analysis, color-coded with green accent
- Decisions — collapsible items with situation/choice/reasoning, color-coded with blue accent
Prompt Quality tab
- Dimension scores — five 0–100 scores for context_provision, request_specificity, scope_management, information_timing, and correction_quality
- Score badge — overall score shown in the tab header for quick scanning
- Deficit findings — categorized prompting problems with before/after takeaways
- Strength findings — effective behaviors worth repeating
- If not yet analyzed, shows an inline Analyze button
Conversation tab
- All messages in chronological order
- Tool calls shown inline with parameters and results
- Thinking blocks (Claude Code only)
- Code blocks with syntax highlighting
- Context break dividers marking where the conversation context was compacted
- Inline command chips for slash commands used during the session
- Raw message toggle to view unformatted message content
Searching and filtering
Use the search bar to find sessions by title. Filter by:
- Project — select from the left panel (grouped by git remote or path)
- Character — session type (Deep Focus, Bug Hunt, etc.)
- Status — analyzed or not analyzed
- Source tool — Claude Code, Cursor, Codex CLI, Copilot CLI, VS Code Copilot Chat (in the project nav panel)
Sessions are grouped by date (Today, Yesterday, This Week, etc.) with daily separators.
Renaming sessions
The auto-generated title can be customized. Click the pencil icon next to the title in the detail view to rename it. When analysis is run, the AI may also suggest a better title — you can apply or dismiss the suggestion.