bobbin-ai 0.25.1

Bobbin gives developers and coding agents search and context over their repositories. Use its command line, Model Context Protocol (MCP) server, or agent hooks to find code by meaning, search exact words, and discover files that change together. Local indexing and search need no API key.

Why you would want it

• Find relevant functions even when you do not know their names.
• Give your agent a context bundle that fits a line or token budget.
• See related files that imports alone do not reveal, using Git history.

Read how Bobbin combines these signals.

Install

Checksummed releases support Linux and macOS on x86-64 and ARM64. For Linux x86-64, with curl, tar and sha256sum installed:

mkdir -p bobbin-download && cd bobbin-download
base=https://github.com/scbrown/bobbin/releases/download/v0.18.1
asset=bobbin-v0.18.1-x86_64-unknown-linux-gnu.tar.gz
curl -fLO "$base/$asset" && curl -fLO "$base/SHA256SUMS.txt"
sha256sum --check --ignore-missing SHA256SUMS.txt && tar xzf "$asset"
export PATH="$PWD/bobbin-v0.18.1-x86_64-unknown-linux-gnu:$PATH"
bobbin --version

Keep the extracted directory intact: its lib/ contains ONNX Runtime. The pinned release reports:

bobbin 0.18.1 (bc246f84003f89eec07618e8c341bb6835500f58)

For other platforms and runtime prerequisites, use the matching release asset. With the Rust build prerequisites installed, cargo install bobbin-ai --locked builds from source and installs bobbin. Run bobbin --version; if it reports an older version, check command -v bobbin for another installation earlier in PATH.

First success in three commands

Run in a shell with Python 3 and Git installed. This creates a disposable project; initial indexing downloads the embedding model and runs locally on the CPU.

mkdir -p bobbin-demo && cd bobbin-demo && git init -q && printf 'def greeting(name):\n    return "Hello, " + name\n' > greeting.py
bobbin init --quiet && BOBBIN_GPU=0 bobbin index --quiet --skip-calibrate
bobbin grep greeting --json | python3 -c 'import json,sys; r=json.load(sys.stdin)["results"][0]; print("{}:{}-{}".format(r["name"], r["start_line"], r["end_line"]))'

The final command prints:

greeting:1-2

Bobbin found a parsed function, with its source range, in the index you just built. See the quick start for semantic search.

On your own code

Run bobbin init and bobbin index from your repository first.

See the CLI reference for flags and additional commands.

Wire it into your agent

From the indexed repository, bobbin serve starts an MCP server on standard input/output. Configure your MCP client to run it in that repository:

{"mcpServers":{"bobbin":{"command":"bobbin","args":["serve"]}}}

For automatic Claude Code context injection, run bobbin hook install in the repository; it updates that project's .claude/settings.json. See agent setup for client configuration and hooks for gating, deduplication and removal. Prefer MCP when available, then the CLI; the HTTP API is the transport fallback.

Before you start

See installation for system libraries and source-build requirements, and language support.

What's next

• Read the Bobbin book.
• Find every document in the docs map.
• Build useful context for a task.

🧺 The stack

Caboodle installs these together and proves each one works; every tool also stands alone.

Contributing

just build
just test
just check

See CONTRIBUTING.md for prerequisites and documentation checks.

📜 License

Apache License, Version 2.0. Releases before 2026-09-24 were MIT-licensed.