Kungfu UNGFU™
One human, agents, and 3,467 merged PRs in 30 days
Last checked:
What is Kungfu UNGFU™?
Kungfu is a platform for maintaining work continuity across different AI agents. It enables the same task to persist across Claude, Codex, or other execution surfaces without context loss, separating work management from agent selection and providing auditable evidence of agent handoffs, failure recovery, and completion workflows.
Kungfu UNGFU™ pricing
Pricing model: Freemium
Kungfu UNGFU™ pros
- Agent-agnostic: seamlessly switches between different AI agents (Claude, Codex, OpenCode) while keeping work state intact
- Failure-resilient: Work survives connection drops and process crashes, retained as Attempts under continuous Work tracking
- Governed completion: separates Agent execution from independent review and settlement approval, preventing agents from approving their own results
- Evidence-based: all claims backed by auditable video proofs, transcripts, and passports for transparency and verification
Kungfu UNGFU™ cons
- Public alpha status: early-stage software with stability and feature completeness limitations
- Steep learning curve: complex conceptual model (KFD, Agent Supply Chain, Hubs) may be difficult for users unfamiliar with agent architectures
- Limited ecosystem: currently lacks multi-Hub market and external vendor adoption; not yet production-ready
Frequently asked questions about Kungfu UNGFU™
Can my work survive when switching to a different agent?
Yes. Kungfu keeps the same work moving across fresh agent sessions without copying chat context or re-explaining decisions.
What happens if an agent crashes or the connection drops?
The work survives failures; dropped connections and crashed processes are retained as Attempts under the same Work, resumable by the next agent.
Can an agent approve its own completed work?
No. Agents produce candidates and evidence, but Kungfu settlement enforces that only independent review can approve and complete the work.
Which AI agents does Kungfu support?
Kungfu is agent-agnostic and works with Codex, Claude, OpenCode, or any custom execution surface, letting you choose the best agent for each task.