Trail-Code
Trail-Code describes itself as a way to turbocharge Claude Code, Cursor, Codex, Gemini and other coding agents with contextual understanding of your codebase. Its README has two parts. The first, Trail, targets repeated corrections: when you correct an agent, the lesson becomes a rule kept in your CLAUDE.md and AGENTS.md, and rules that must not be broken can get a blocking hook. Learnings that agents save after a session go into `.trail/learnings/` in the repo, are committed with the change and are read back by the next session, yours or a teammate's. `trail check` compares a change with what the team has learned before a PR.
The second part, Graft, builds a graph of your code as linked markdown files, one node per system, API or concept, saved in a `graft/` folder that is git-ignored like a local cache. The README says it rebuilds against the working tree on every query, needs no embeddings or server, and can be wired into Claude Code and other agents, including as an MCP server. Install with `npm install -g @trailhq/trail` and `trail init`; the Graft section uses the `@nanonets/graft` package name.
The README leads with numbers such as up to 4x cheaper and 3x faster, a 42% token saving and correctness rising from 54% to 66%. These are the project's own benchmark figures, and the README mixes Trail and Graft naming, so read the benchmark section and the packages you actually install carefully. A team dashboard needs signing in to a hosted service. It is MIT licensed, created in July 2026, and was last pushed on 2026-10-13. This review is based on the README and repository data; we have not run it ourselves.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-11-06
- Commits, last 90 days
- 100+
- Releases, last 90 days
- 0
- Latest release
- —
- Open issues
- 289
- Recent issues with at least one comment
- 0/6
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #374 Runaway upkeep child processes when ~/.graft is unwritable in a sandbox5 comments · 2026-09-13
- #323 0.13.0+ won't start on Windows without a C toolchain: [email protected] ships no prebuilds, so require() throws before any command ru5 comments · 2026-09-09
- #326 Using skill instead of AGENTS.md4 comments · 2026-09-09
- #313 Do not clobber statusline on first install4 comments · 2026-09-09
- #298 graft build --deep: hardcoded maxTokens=8192 truncates tool-call JSON, and file-node summaries discarded due to id-decoration mismatch4 comments · 2026-09-04
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Who it fits
- You keep correcting your agent the same way and want each lesson saved as a rule in CLAUDE.md and AGENTS.md.
- You want learnings committed in `.trail/learnings/` so the next session or a teammate gets them back.
- You want a code graph as linked markdown files, with no embeddings or server.
- You need independent benchmarks: the 4x cheaper and 3x faster claims are the project's own.
- You want team views without a hosted account: they need sign-in to a hosted service.
- You prefer mature tools: the repository was created in July 2026.
How it differs
Context Mode keeps tool output out of the window during a session. Trail-Code carries lessons between sessions through files in the repo.
Compatibility
Quick start
Use cases
Alternatives
Featured in these lists
Comments
No comments yet. Be the first!