BoostGraph

How it works

BoostGraph reads your codebase and builds a local map of how it fits together — what calls what, how files connect, and where key logic lives — so your agent can answer structural questions without searching file by file.

How agents query it

BoostGraph runs as an MCP server and exposes a tool that lets a supported coding agent query your codebase with natural-language questions — no manual file search required.

Seamless indexing

Indexing happens in the background. When you run boost init, Boost registers hooks on your supported coding agents. On each session start, the hook verifies the index:

• First time in a repo: creates .boost/ and builds the graph quietly while you work
• Later sessions: syncs the index to match the current tree
• After code changes: a local file watcher keeps the index updated

On large codebases, the first index can take longer — but it runs in the background and does not interfere with your agent. When the index is ready, the agent can use boostgraph_explore and you get the full benefits of BoostGraph.

To index manually ahead of time, run boost graph init in your repository directory.

How to use BoostGraph

1. Run boost init once to wire Boost into your agents (session-start hooks, MCP config). Restart your agent afterward.
2. Open your codebase in the agent — indexing starts automatically in the background. On large repos, wait for the first index to finish or pre-build with boost graph init.
3. Ask structural questions — flows, architecture, callers/callees. The agent should reach for boostgraph_explore instead of long grep/read loops.

Benefits

• Faster answers — one to a few tool calls instead of dozens of discovery steps
• Surgical context — exact symbols and call paths, not whole directory trees
• Lower cost — less discovery overhead means fewer tokens per session
• 100% local — no API keys, no data leaves your machine
• 30+ languages — Go, Python, TypeScript, Java, Rust, C#, and more

Demo video

Side-by-side comparison: grep-and-read exploration vs. BoostGraph.

Benchmark: Minikube runtime preflight flow

Agent performance with vs without BoostGraph on a deep flow-tracing question in kubernetes/minikube:

What happens internally when Minikube attempts to create a cluster using a container runtime that is not installed or enabled on the host? Trace preflight validation, missing-runtime detection, automatic fallback or download, and how the error is formatted before exit.

Troubleshooting

BoostGraph not initialized

Run boost init so session-start hooks are registered, then open the project with your coding agent — indexing starts automatically. For immediate indexing, run boost graph init in the repo root.

Indexing is slow

First-time indexing on large repos takes longer; it runs in the background and does not block your agent.

Agent still greps instead of using BoostGraph

The index may still be building, or MCP may not be connected yet because indexing is still in progress. Restart your agent and verify MCP is connected via /mcp.