AgentIndex icon
AgentIndex
ai-devkit logo
ai-devkit
★ 1.7k
vs
kungfu logo
kungfu
★ 4.5k

ai-devkit vs kungfu

ai-devkit: AI DevKit is a local-first operating layer for teams running several coding agents. One .ai-devkit.json reconciles skills, MCP and setup across Claude Code, Codex CLI, Gemini CLI, opencode, Pi, Cursor, Copilot and others; agent console is a live TUI for running sessions; agent send routes prompts, logs or stdin to a running agent; and a local SQLite memory is searched over MCP when needed. A dev-lifecycle skill guides requirements, design, planning, implementation, testing and review.; kungfu: Kungfu keeps a unit of Work outside any single agent session, recording its objective, progress, evidence, next action and completion state, so the same Work can move between Codex, Claude, OpenCode, Amp or your own agent without re-explaining context. Each Agent attempt is recorded under the Work, a second live writer is blocked, and an Agent cannot approve its own result: completion goes through independent review. It is in Alpha.

01

TL;DR

ai-devkit logo
ai-devkitDev Tooling · ★ 1.7k

Keeping one setup across Claude Code, Codex and Cursor in a team repo

Read ai-devkit full review →
kungfu logo
kungfuDev Tooling · ★ 4.5k

Switching between a premium and a cheaper coding agent on the same task

Read kungfu full review →
02

Side-by-Side Comparison

Field
ai-devkit logoai-devkit
kungfu logokungfu
Category
Dev Tooling
Dev Tooling
Stars
★ 1.7k
★ 4.5k
License
Apache-2.0
Apache-2.0
Install Command
npx ai-devkit@latest setup && npx ai-devkit@latest init
curl -fsSL https://kungfu.tech/install.sh | sh
Updated
1d ago
1d ago
Open Source
Yes
Yes
Website
↗ Visit
↗ Visit
GitHub
↗ GitHub
↗ GitHub
Tags
coding-agents, claude-code, codex
agent-handoff, durable-work, codex
03

Features

ai-devkit logoai-devkit
01One .ai-devkit.json that writes agent files, skills and MCP setup for each tool you pick
02agent list and agent console to supervise sessions across providers
03agent send with --stdin, --wait and groups to route work to running agents
04Local SQLite memory exposed via MCP and CLI, retrieved only when useful
05dev-lifecycle, verify, tdd, review, debugging and security skills
kungfu logokungfu
01Project, Work and Attempt model that keeps durable state outside the chat
02kungfu run codex|claude|opencode|amp starts an agent with its native console while Kungfu tracks the Work
03Recovers the same Work after disconnects and crashes, with a deterministic mock-agent demo
04Blocks a second live agent from writing to a Work it does not own
05Completion requires independent review and settlement, not the agent's own claim
04

Use Cases

ai-devkit logoai-devkit
↳Keeping one setup across Claude Code, Codex and Cursor in a team repo
↳Piping test output to a running agent or a group of reviewer agents
↳Giving agents searchable project conventions without bloating every prompt
kungfu logokungfu
↳Switching between a premium and a cheaper coding agent on the same task
↳Keeping long-running work resumable across context loss and restarts
↳Making agent work auditable with evidence and a reviewed completion step
FAQ

FAQ

What is the difference between ai-devkit and kungfu?
Both ai-devkit and kungfu are in the Dev Tooling category. ai-devkit has 1.7k stars, while kungfu has 4.5k stars.
Which is better, ai-devkit or kungfu?
The best choice depends on your use case. Choose ai-devkit if Keeping one setup across Claude Code, Codex and Cursor in a team repo, and kungfu if Switching between a premium and a cheaper coding agent on the same task.
Is ai-devkit free or open source?
Yes, ai-devkit is open source on GitHub (Apache-2.0).
Is kungfu free or open source?
Yes, kungfu is open source on GitHub (Apache-2.0).
Cloud Infrastructure
Ad

Deploy on DigitalOcean

Host your MCP servers, Docker containers, and AI agents with $200 free credit over 60 days.

Claim $200 Free Credit
→

Related

Alternatives to ai-devkit →Alternatives to kungfu →ai-devkit details →kungfu details →