Analytics & Export
View activity charts and cost analytics. Export session insights with quick Markdown downloads or AI-powered synthesis across sessions.
Analytics dashboard
The Analytics page provides charts of your coding activity:
- Activity over time — sessions per day or week
- Model usage — sessions and messages by model
- Session characters — distribution of session types (deep focus, bug hunt, feature build, etc.)
- Cost trends — spending over time by project and model
- Tool call patterns — which tools you use most
All charts respond to the period selector: 7 days, 30 days, 90 days, or all time.
Quick session export
Export a single session directly from the session detail panel.
- Open the dashboard and navigate to Sessions
- Click into any session to open the detail view
- Click the download icon (↓) in the panel header
- Choose a format from the dropdown:
| Format | Output |
|---|---|
| Markdown | Standard .md with session metadata, summary, and insights |
| Obsidian | Markdown with YAML frontmatter for Obsidian vaults |
| Notion | Markdown formatted for Notion import |
The export includes the session title, date, duration, project, summary, and all extracted insights (decisions, learnings, techniques, prompt quality).
Run analysis on a session first for richer exports. Without analysis, the export contains only basic session metadata.
AI-powered export
The Export page uses your configured LLM to synthesize insights across multiple sessions into curated, deduplicated knowledge. Unlike quick exports, the AI compiles knowledge — it doesn't just concatenate raw data.
AI-powered export requires an LLM provider configured in Settings. Supports OpenAI, Anthropic, Gemini, and Ollama.
The 4-step wizard
Choose scope
Select what to include:
- Project — insights from a single project. All output rules are implicitly scoped to that project.
- All projects — insights across every project. The AI labels each rule as UNIVERSAL (applies everywhere) or PROJECT-SPECIFIC (applies only in a certain context).
Configure format and depth
Pick an output format and a depth preset. A stat bar shows how many insights are available for your selected scope.
Generate
Click Generate with AI. Progress streams in real time:
- Loading insights from the database
- Sending to your LLM for synthesis
- Generation complete — auto-advances to the next step
You can cancel mid-generation to save tokens.
Review and export
Preview the generated content, then:
- Copy to clipboard — paste directly into your editor
- Download as .md — saves a Markdown file
Click Start over to try a different format or scope.
Output formats
| Format | Best for | What it generates |
|---|---|---|
| Agent Rules | CLAUDE.md, .cursorrules, agent configs | Imperative instructions with WHEN conditions and REVISIT triggers. Includes a "Prompt Hygiene" section. |
| Knowledge Brief | Documentation, wikis, team onboarding | Structured markdown with executive summary, decisions, learnings, and techniques organized by topic. |
| Obsidian | Obsidian vaults | Markdown with YAML frontmatter and [[wikilinks]] for cross-referencing between notes. |
| Notion | Notion pages | Markdown with toggle blocks, callout blocks, and tables — optimized for Notion's block model. |
Depth presets
| Depth | Insights included | Use case |
|---|---|---|
| Essential | ~25 top insights | Quick reference. Top rules only — fast and focused. |
| Standard (default) | ~80 insights | Key decisions, learnings, and techniques. Good balance of coverage and conciseness. |
| Comprehensive | ~200 insights | Everything available. Most thorough — useful for full project knowledge dumps. |
What the AI does
The LLM doesn't just reformat — it compiles knowledge:
- Deduplicates overlapping learnings across sessions
- Resolves conflicts between decisions made at different times (notes the evolution)
- Prioritizes by frequency and confidence score
- Contextualizes each rule with WHEN conditions for applicability
- Classifies scope (in All mode) — labels rules as UNIVERSAL or PROJECT-SPECIFIC
Example: Agent Rules output
## Architecture
- ALWAYS use SQLite with WAL mode for local-first applications
- WHEN: Building CLI tools or desktop apps that need persistent storage
- REVISIT WHEN: Multi-device sync becomes a requirement
- PREFER pnpm workspaces for monorepo management over npm/yarn
- WHEN: Project has 2+ packages with shared dependencies
## Prompt Hygiene
- PROVIDE file paths and line numbers when asking for code changes
- BAD: "Fix the bug in the auth module"
- GOOD: "Fix the null check in src/auth/middleware.ts:45"