# kungfu-systems/kungfu

Your agents don’t hand off the work. Kungfu keeps the same Work moving across Codex, Claude, OpenCode, and your own execution surface.

Repository: https://github.com/kungfu-systems/kungfu
Canonical: https://ross.abutalabs.com/products/kungfu
Homepage: https://kungfu.tech
Language: C++
License: Apache-2.0
License Family: permissive
Topics: agent, agent-framework, agent-memory, agent-orchestration
Last push: 2026-08-26T23:13:46+00:00

## Health v2 (maintenance only)
Score: 99/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 99, longevity 100
- inputs: {"age_days": 3213, "days_push": 7, "days_rel": 9, "gap_med": 0, "n_releases_24m": 4}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4502, forks 1286 (observed 2026-08-28T04:08:51.531194+00:00)

## What it is
Kungfu is an open-source continuity layer for AI agent work, consisting of a local C++ runtime (libkungfu) and an open protocol (KFD) that let a unit of Work keep its identity, state, and evidence as it moves across agents like Codex, Claude, OpenCode, and Amp. It ships as a standalone CLI plus optional TUI/GUI sidecar views, with durable project-local state in a .kungfu/ workspace.

## Use cases
- hand off an in-progress task from one AI coding agent to another without copying context
- resume agent work after a crash or dropped connection
- switch from an expensive model to a cheaper one mid-task while keeping the same Work
- audit and inspect what an agent did across multiple sessions
- run a deterministic mock-agent recovery scenario to test continuity locally
- keep durable, machine-readable records of agent attempts, approvals, and next actions

## When to choose
- you regularly switch between multiple AI coding agents (Codex, Claude, OpenCode, Amp) on the same project
- you need work to survive agent crashes, disconnects, or vendor/model changes
- you want a local, auditable continuity layer without a mandatory cloud or vendor lock-in
- you want to mix expensive and cheap agents on the same unit of work

## When to avoid
- you use a single agent exclusively and never hand off or resume work
- you need a production-stable, generally available tool - it is currently in public Alpha
- you want a fully autonomous multi-agent framework rather than a work-continuity and handoff layer
- you cannot tolerate a project-local .kungfu/ workspace directory in your repos

## Facets
- artifact type: cli-tool
- maturity: experimental
- function: agent-framework, workflow-automation, developer-tools, cli
- domain: developer-tools
- platform: windows, cli, cross-platform
- tags: agent-orchestration, agent-continuity, work-handoff, agent-memory, local-runtime, open-protocol, multi-agent, ai-agents, automation, command-line, macos, linux

## Member repositories
- kungfu-systems/kungfu (main) score 99

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:51.531194+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:20:24.576145+00:00, confidence not recorded.
  - readme: https://github.com/kungfu-systems/kungfu (fetched 2026-08-28T04:08:51.531194+00:00, sha b1cda20d61c0)
  - homepage: https://kungfu.tech (fetched 2026-08-29T09:06:29.841773+00:00, sha 6f103319854b)
  - site_page: https://kungfu.tech/about/index.html (fetched 2026-08-29T09:06:29.850708+00:00, sha d80019bb7162)
- Data as of 2026-08-30T08:39:29.467469+00:00.