Analysis updated 2026-05-18
Let a fresh AI agent session continue project work without re-explaining context.
Surface missing or conflicting project understanding instead of letting an agent guess.
Record durable work facts that other tools can read across a shared boundary.
Build the current source tree to evaluate early functionality before a public release.
| kungfu-systems/kungfu | skeeto/w64devkit | ultrajson/ultrajson | |
|---|---|---|---|
| Stars | 4,351 | 4,481 | 4,482 |
| Language | C++ | C++ | C++ |
| Setup difficulty | hard | easy | easy |
| Complexity | 4/5 | 2/5 | 2/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
No public release artifact yet, only a source build path is available, and features are still in progress.
Kungfu addresses a problem familiar to anyone using an AI coding agent: when a chat session ends, the agent loses its understanding of the project and the person has to explain everything again. Kungfu aims to let a fresh agent pick up the same work without that repeated explanation, by recording project sources, surfacing missing or conflicting information instead of letting the agent guess, and keeping enough structure around to continue the work later. The project describes itself as the foundation of something it calls an Agent Supply Chain, a layered system where Kungfu is responsible for recording, ordering, checking, and recovering durable facts about work in progress, which the project calls Facts and Episodes. Other tools and products can then read this durable state across a boundary, rather than each one keeping its own private, disconnected memory of a project. The intended everyday experience is a single command, kungfu run agent, run from inside a project folder, after which the agent explains what it understands about the project, points out anything unclear or conflicting, and keeps that understanding consistent across restarts and different agent tools. The README is careful to say this command describes the intended first release experience rather than something available as a finished public download today. Status is explicitly listed as coming soon, with a v4 alpha that the maintainers describe as evidence only, not yet carrying the stability guarantees planned for a full release. The README also discusses how the project plans to handle version upgrades and users leaving the platform, treating both as commitments the software should honor rather than afterthoughts. People wanting to try the current source code can clone the repository and use included build scripts, though the maintainers note that public, ready to use release artifacts are not yet available. The project is written mostly in C++ and is licensed under Apache 2.0.
A tool that lets AI coding agents remember and continue project work across separate chat sessions.
Mainly C++. The stack also includes C++.
You can use, modify, and share this freely, including commercially, as long as you keep required notices and follow the patent terms.
Setup difficulty is rated hard, with roughly 1day+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.