Open-source foundations

Fehm combines its own repository graph with selected open-source components. Context improvements run through your existing commands. Model-based review is an optional workflow you configure separately.

Context that spans relevant files

Graft contributes a file-selection algorithm. Fehm keeps the three strongest ranked results first, then rotates the remaining candidates across files. This helps reduce repeated-file crowding within a context budget. Small budgets can still omit files; search scores remain unchanged.

fehm scan .
fehm context "change authentication" --budget 4000

Find code and relevant decisions

Codebase Memory contributes identifier-boundary logic, helping retrieval recognize names such as HTTPServer and XMLParser. Everything Claude Code contributes relevance-ranking logic that prioritizes stored decisions and warnings matching your question or detected technologies. Neither requires a separate installation.

Review changes with source context

Open Code Review is bundled in Fehm’s platform-specific wheel. Configure your model provider, preview a review, then explicitly allow provider execution. Fehm can supply a context packet and link returned findings to indexed source locations.

Agency Agents contributes the optional code-review guidance profile, covering correctness, security, maintainability, performance and tests. gstack contributes the source-freshness classification used to report whether indexed source changed during review or was already stale.

fehm review --context "authentication changes" --profile code-review --graph-report

This command previews the plan. Follow the Open Code Review guide to configure and run it. Findings remain model-generated claims. Freshness checks cover indexed files, not new files outside the index, and cannot detect files changed and restored between snapshots.

What you install

Your Fehm wheel includes these selected adaptations and their license notices. You do not need to install the five source projects separately. They are not additional assistant integrations or complete installations of those products. Ordinary graph queries remain local; configured model review can send source to your provider.

Project credits

Selected source adaptations use MIT licenses: Graft, Everything Claude Code, gstack, Agency Agents and Codebase Memory. The bundled Open Code Review engine uses Apache-2.0. Pinned source manifests and license notices ship with Fehm.