Introduction

Code Insights parses your Claude Code session history — the JSONL files stored in ~/.claude/projects/ — and syncs them to your own Firebase database. From there, a web dashboard lets you browse sessions, generate LLM-powered insights, and track patterns in how you work with AI.

Why?

Claude Code stores every conversation as JSONL. That’s valuable data: what you built, why you made certain choices, what went wrong and how you fixed it. But raw JSON isn’t searchable or browsable. Code Insights makes it useful.

Privacy: Bring Your Own Firebase

There is no central Code Insights server storing your data. The architecture is simple:

What	Where	Who can access
Your session data	Your Firebase	Only you
Login credentials	Supabase Auth	Authentication only
Analytics	Vercel Analytics	Aggregate, anonymous

You create a Firebase project, point the CLI at it, and your data stays there. The hosted dashboard at code-insights.app connects to your Firestore to display it — it never stores your session data.

How It Works

~/.claude/projects/**/*.jsonl
           |
           v
    +--------------+
    |   CLI         |  Parse JSONL, extract metadata
    |  (Node.js)    |  Upload to YOUR Firestore
    +--------------+
           |
           v
    +--------------+
    |  Firestore    |  projects, sessions, messages
    |  (YOUR DB)    |  <- You own this data
    +--------------+
           |
           v
    +------------------------------------+
    |  Web Dashboard (code-insights.app) |
    |  Reads from YOUR Firestore         |
    |  LLM insights with YOUR API key    |
    +------------------------------------+

The system has two parts:

• CLI (this repo, open source) — Parses JSONL files, generates session titles, classifies session types, and syncs everything to Firestore.
• Web Dashboard (closed source, hosted at Vercel) — Visualizes sessions, runs LLM analysis, and exports data to Markdown.

What You Get

• Session browser — Search, filter, and read full session transcripts.
• Smart titles — Sessions are automatically titled based on content, so you can find things later.
• Session classification — Each session gets a character type: deep focus, bug hunt, feature build, exploration, refactor, learning, or quick task.
• LLM analysis — Generate summaries, decisions, learnings, and technique insights using your own API key (OpenAI, Anthropic, Gemini, or Ollama).
• Multi-device sync — Sync from multiple machines to the same Firebase. Project IDs are stable across devices when using git.
• Markdown export — Download insights in plain, Obsidian, or Notion format.

Next Steps

Ready to set it up? Head to Installation.