Graft: Open-Source Tool Turns Claude Code's Context Into Shared, Persistent Docs

The biggest problem with Claude Code has just been solved.
Everything your agent learns from your code gets wiped out when you close the session.
And tomorrow it learns it all over again. And your teammate learns it from scratch again with their own agent. And again. And again.
Nobody on your team is saving that work anywhere.
This turns it into files:
→ Generates a repo map in markdown
→ Saves to your project
→ Updates only when the code changes
→ You share it via git and it benefits the whole team
→ Runs locally, no accounts or keys needed
A new hire reads the docs once. Your agent hasn't read anything in months.
It's called Graft. Open source, MIT license.
Link: https://lnkd.in/eNhr5f7z

Solid check here — Graft is real, and the description tracks closely to the project's own marketing copy (in fact, this post reads like a close paraphrase of Graft's own README/landing page language). Details:

Confirmed accurate:

• Graft is real, open source, MIT licensed, built by NanoNets — confirmed via GitHub, npm, and the official site (graft.nanonets.ai).
• Generates a repo map as linked markdown files, written into your project — confirmed exactly: "Graft builds that understanding once and writes it into your repo as a folder of linked markdown files, one node per system, API, or concept."
• Updates when code changes — confirmed via its "auto-sync" and background-rebuild mechanism.
• Runs locally, no accounts or API keys needed — confirmed ("Deterministic, no key").
• Plugs into Claude Code specifically (via hooks), also supports Cursor, Codex, Gemini — confirmed.
• The "agent relearns your codebase every session, nobody saves the work" problem framing — this is almost a direct paraphrase of Graft's own pitch: "Every task, your coding agent starts blind... Repeated... Discarded... Unshared." So the post's argument isn't independently fact-checkable so much as it's restating the vendor's own framing as if it were a neutral observation — worth knowing if you want to distinguish your own analysis from the project's marketing copy.

One nuance worth flagging:

• The post implies you generate the markdown once, commit it, and teammates get the benefit automatically via git. The project's own README is slightly more specific: what actually gets shared via git is the wiring (config dropped into .claude/), and each teammate still runs graft build locally to generate their own graph. So "you share it via git and it benefits the whole team" is directionally true but slightly oversimplified — teammates don't just passively inherit a finished map, they still run a build step.

Not independently verifiable / worth caution:

• "The biggest problem with Claude Code has just been solved" — this is marketing hyperbole, not a checkable claim. Context loss between sessions is a known pain point, but "the biggest problem" and "solved" are subjective framing, not facts.
• A third-party skills directory (SkillsLLM) notes that Graft "returned warnings in SkillsLLM's automated security scan." I couldn't verify what specifically triggered that, but it's worth investigating before amplifying the tool uncritically, especially since it runs locally with repo access.
• I also found near-identical forks of the repo under other GitHub usernames (RShuken, adrianlerer) with the same README — likely just forks/clones of the original NanoNets project rather than evidence of anything wrong, but worth being aware the ecosystem around it is a bit noisy.

Bottom line: the factual/technical claims in the post are accurate and match the project's own documentation closely. The main things to soften are the "biggest problem, just solved" hyperbole and the slightly oversimplified "just share via git" framing.