🚀 Roadmap to v1.0

[CodewithJha]

Current Status

Mutiny is an open-source behavioral fuzz-testing engine for AI agents: policy-as-code on tool-call traces, evolutionary campaigns, minimize, and permanent regressions. Adapter #1 (OpenAI Agents SDK), CLI, Hosted UI, and a solid test suite are in place. Install from source today (uv sync --extra dev) — not on PyPI yet.

Priorities may change with community feedback. Use Discussions to brainstorm; open Issues when work is actionable.

✅ Completed

• Core — Framework-independent policy oracle, campaign search, minimize, regression replay
• Hosted UI — Optional lineage / evidence surface (apps/web, apps/api)
• OpenAI Agents SDK Adapter — First shipped TargetAdapter
• CLI — mutiny init / run / test into your agent project
• Policies — Deterministic policy.yaml operators (code proves, not LLM-as-judge)
• Campaigns — Evolutionary behavioral fuzzing against live agent paths
• Tests — Unit, integration, and reliability coverage under tests/
• Regressions — Frozen findings under .mutiny/tests/ replayed with mutiny test

🛣 Next

• LangGraph adapter (on the existing TargetAdapter port)
• CrewAI adapter
• UI polish — Hosted screenshots, a11y, clearer empty states
• GitHub Action — CI-friendly mutiny test / campaign hooks
• PyPI — Publish installable packages
• More examples — Policy packs and sample projects beyond Adapter docs: add policy.yaml operator cheatsheet #1

🌎 Future

• Cloud Hosted — Managed campaign / evidence surface
• Auth — Multi-user access for Hosted
• Teams — Shared policies, findings, and regressions
• CI integrations — Deeper GitHub / GitLab / other CI wiring
• More adapters — PydanticAI, AutoGen, HTTP, and community-proposed frameworks

How to help

• Newcomers: pick a good first issue
• Ideas & brainstorming: Discussions
• Contributing guide: CONTRIBUTING.md
• Living doc: docs/ROADMAP.md