✦ Open source · MCP server · Python

The why behind
your AI-written code

A git blame for AI agents — but for the why, not just which line which model touched. Captured live, by the agent, in the moment. So six months later you still know what each schema change, env var, and dependency was actually for.

View on GitHub Get started →
pip install selvedge Click to copy
bash
$ selvedge blame payments.stripe_customer_id ───────────────────────────────────────────────── entity payments.stripe_customer_id changed 2025-11-03 14:22:01 type add agent claude-code commit a3f9c12 diff + stripe_customer_id TEXT   reasoning Links to Stripe customer object for recurring billing. Stored as TEXT to match Stripe's cus_ prefix format. Added per user request for subscription billing integration.
The problem

Git tracks what. Selvedge tracks why.

With human-written code, intent leaked into commit messages and PR descriptions. With AI-written code, intent lives in a prompt — and vanishes when the session ends.

Without Selvedge
✗ "Why does this column exist?" — no one knows
✗ Git blame shows a commit from an AI agent with a generated message
✗ The reasoning lived in a prompt that expired 3 months ago
✗ Schema decisions get re-litigated endlessly
With Selvedge
✓ Every change has captured reasoning from the agent session
✓ selvedge blame users.email answers in seconds
✓ Search by keyword across all reasoning, ever
✓ History survives agent upgrades, team changes, time
How it works

Set up in 5 minutes.
Query forever.

Selvedge runs as an MCP server. Your AI agent calls it as it works. You query it whenever you need answers.

1
Install & init
Install the package and initialize a local SQLite database in your project root.
2
Connect your agent
Add Selvedge to your Claude Code MCP config. Tell your agent to call log_change as it works.
3
History accumulates
Every schema change, function, endpoint, and dependency gets logged with full context.
4
Query anytime
Use the CLI or MCP tools to answer any question about your codebase's history.
# 1. Install pip install selvedge && selvedge init # 2. Add to ~/.claude/config.json { "mcpServers": { "selvedge": { "command": "selvedge-server" } } } # 3. Query selvedge diff users # full table history selvedge blame payments.amount # who, when, why selvedge history --since 30d # last 30 days selvedge search "stripe" # full-text search
Features

Everything your team needs
to stay in sync.

Five MCP tools. A full CLI. Zero config. Works with any AI agent that supports MCP.

🔍
Semantic blame
Not just who wrote the line — which agent, which session, what the user asked for, and why the decision was made.
selvedge blame
📜
Prefix history
Query an entity or a prefix. users returns the full table, all columns, everything that ever touched it.
selvedge diff
🔎
Full-text search
Search across entity paths, diffs, reasoning, and agents. Find everything related to a feature, decision, or bug.
selvedge search
⚡
MCP native
Five tools AI agents call directly: log_change, diff, blame, history, search. Works with Claude Code, Cursor, and any MCP client.
MCP server
🗄️
Zero config SQLite
Local SQLite database. Commit it to share history with your team, or keep it local. No servers, no accounts required.
pip install selvedge
🧩
Structured events
Every event captures entity path, change type, diff, reasoning, agent, session ID, and git commit. Queryable, filterable, exportable.
--json flag
Get started

Start capturing history today.

Open source. Free forever. One command to install.

View on GitHub Read the docs →
pip install selvedge