AEndrix03/Graft
AEndrix03/Graft has 70 GitHub stars and 7 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-17 02:54 UTC. Last push 2026-11-14.
Measured star growth
- in 24h
- 0
- in 7 days
- Collecting history
- in 15 days
- Collecting history
- in 1 month
- Collecting history
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
About
Your coding agents forget. Graft doesn't. Persistent local memory that brings back fixes, decisions and project knowledge when they matter — across Claude Code, Codex, ChatGPT and more. Local-first. No SaaS. No API key.
Developer ToolsREADME.md
OriginalHow it works

Your agent writes down what it learns: the fix, the gotcha, the decision and why. Next time it runs into the same thing, graft hands it the answer before it starts over. The agent spends a moment checking it still holds, then moves on.
- Fewer tokens. No more whole turns of reasoning for a problem already solved once.
- Faster. A lookup takes a fraction of a second.
- Consistent. The same problem gets the same answer, not a new invention every session.
Stop solving the same problem twice
without graft bug → investigate → fix → forgotten (again, next week)
with graft bug → recall → check → done
Benchmark
- It stops making things up. Without graft, a quarter of the answers about the team's own code were invented. With graft, none.
- A wrong note does not derail it. On 30 questions no note answers, graft sometimes offered the wrong note, and the agent set it aside every time.
- Where it still slips. graft finds the right note for 88% of the questions. When it picks a wrong one, the question is usually about the same technology as another note: that is the next thing being fixed.
How it was measured, and how to run it yourself → bench/
Install
1. The CLI
curl -fsSL https://raw.githubusercontent.com/AEndrix03/Graft/master/install.sh | sh
irm https://raw.githubusercontent.com/AEndrix03/Graft/master/install.ps1 | iex
scoop install https://raw.githubusercontent.com/AEndrix03/Graft/master/bucket/graft.json
brew tap AEndrix03/graft https://github.com/AEndrix03/Graft.git && brew install graft
The installer also copies the skills into every agent it finds. Then run
/graft-init inside your agent.
2. Or straight from your agent's marketplace
/plugin marketplace add AEndrix03/Graft
/plugin install graft@graft
/graft:graft-init
codex plugin marketplace add AEndrix03/Graft
codex plugin add graft@graft
then ask Codex to set up graft.
graft-init installs the CLI if it is missing, asks whether graft should be on
everywhere or in this project only, and you are done.
Building from source, GPU builds, other options → docs/install/
See it work
First, the memory is empty:
$ graft query "spring validation nested dto not working"
{
"status": 0,
"result": { "hit": "MISS" }
}
The agent investigates and solves the issue. Save the useful part:
graft insert \
--title "Spring @Valid must also be applied to nested DTO fields" \
--body "Without @Valid on the nested field, validation does not cascade into it." \
--keyword spring-boot \
--keyword validation \
--keyword gotcha
Weeks later, with different wording:
$ graft query "why are constraints inside my nested request object ignored?"
{
"status": 0,
"result": {
"hit": "STRONG",
"title": "Spring @Valid must also be applied to nested DTO fields",
"body": "Without @Valid on the nested field, validation does not cascade into it."
}
}
Different prompt. Same underlying problem.
Graft surfaces the prior learning. The agent decides whether it is useful.
Browse your memory
![]() |
![]() |
| Every memory and how it connects | From one note to the related ones |
graft view opens it in your browser.
Works with your agent
| plugin from the marketplace | |
| plugin from the marketplace | |
| skills, installed by the CLI | |
integrations/gemini-cli/ |
|
MCP · integrations/claude-ai/ |
|
MCP · integrations/chatgpt/ |
|
anything that can run a command · docs/integrations/ |
Project status
Alpha — v0.2.x. The CLI and its JSON output may still change before 1.0.
Working today:
- one local binary and daemon, memory in a single SQLite file on your machine
- multilingual: a question in Italian finds a note written in English
- plugins for Claude Code and Codex, skills for OpenCode, MCP for Claude Desktop and ChatGPT
- set up once, then hands off: the agent bootstraps each new project, searches, fixes stale notes, saves what it learned and keeps the graph tidy on its own
- on Claude Code, recall without asking: the plugin looks up every prompt and hands the agent a close note when there is one
- a graph viewer to browse every memory, and usage stats to see what actually gets reused
- a reproducible benchmark in
bench/
Coming next:
- Finding the right note more often. In the benchmark every answer the agent still got wrong with graft came from a lookup that missed or picked a neighbouring note. Wrong notes it simply sets aside, so the work goes into recall, not into stricter gates.
- Recall without asking beyond Claude Code. The same per-prompt lookup for the other agents, where their hook systems allow it.
- Signed Windows binaries and prebuilt macOS archives.
Documentation
| Getting started | docs/install/ |
| Use cases | docs/use-cases.md |
| Guide: persistent coding-agent memory | docs/use-cases/persistent-memory-for-coding-agents.md |
| Concepts | docs/concepts.md |
| Integrations | docs/integrations/ |
| Architecture | docs/architecture/ |
| CLI | docs/cli/ |
| Retrieval | docs/retrieval/ |
| Storage | docs/storage/ |
| Embeddings | docs/embeddings/ |
| Profiles | docs/profiles/ |
| HTTP API | docs/http-api/ |
Full documentation → docs/
Contributing
git clone https://github.com/AEndrix03/Graft.git
cd Graft
bash scripts/build-from-source.sh
graft stats
Run tests with:
cmake --build build --target test
See CONTRIBUTING.md.
License
You can use, modify, distribute and embed Graft in proprietary projects subject to the license terms.


