CODEBASE INTELLIGENCE, ON YOUR MACHINE
Build a local graph of files, symbols and dependencies. Retrieve source-backed context, inspect change impact and check architecture drift from your terminal or coding assistant.
Workflow illustration · results depend on repository evidence
Bring the context to your coding assistant.
Connect other compatible clients with MCP →BEFORE YOU CHANGE THE CODE
A shared source graph connects what you read, what you change, and what you check next.
Approve the rules that fit your project. Compare current evidence with a saved baseline and review what disappeared.
Follow changed files into affected dependencies, routes, and tests. Use the evidence to decide what deserves a closer look.
These diagrams explain the workflow. Findings support review; they do not guarantee correctness.
EXPLORE THE FEATURES
Browse ten feature groups with commands, examples and the limits of each analysis.
Trace files, symbols, callers, dependencies, and entry points. Find the source behind the explanation.
Explore capability →Retrieve source excerpts, related tests, architecture rules and project history within a context budget.
Explore capability →Compare changes with approved boundaries. Detect when engineering-practice evidence disappears.
Explore capability →Follow a diff through dependencies, inspect its blast radius, and run supported checks.
Explore capability →Connect coverage, dependency risk, API contracts, infrastructure, and runtime evidence.
Explore capability →Version prompts, review tool contracts, compare model behavior, and evaluate changes against held-out cases.
Explore capability →Watch repositories, explore the local cockpit, preserve decisions, and export onboarding notes.
Explore capability →Inspect test quality, ingest measured coverage, and explore mutation testing and historical bug replay.
Explore capability →Review security paths, dependency risk, API contracts, and infrastructure configuration together.
Explore capability →Check graph-backed claims and preserve agent outcomes, recurring mistakes, and resolution history.
Explore capability →INSTALL WITH UV
Install the packaged tool with uv, register your assistant, and scan. The graph stays on your machine. Provider features make external calls only when you configure and invoke them.
Follow the installation guide →$ fehm scan .
$ fehm query "authentication flow"
$ fehm serve .fehm/graph.json --watch
BEFORE YOU START
No. Install a supplied wheel with uv. The package includes the engine and installs a private Node runtime.
Ordinary graph queries run locally and return evidence. Your coding assistant can use that evidence to explain the code. Model evaluation features require an explicitly configured provider.
TypeScript and JavaScript have the deepest analysis. Python, Go, Rust, and Java have shallower declaration, import, and call resolution.
Follow the setup guide, scan a repository you know, and ask about a familiar flow. Then connect your coding assistant to reuse that context.
SCAN · QUERY · REVIEW
fehm