Self-hosted autonomous team platform. Define a project, attach a pipeline of specialist agents, and let yeet2 plan, execute, review, and ship — continuously.
While the alpha focuses on software development, the architecture is domain-agnostic. Any multi-stage knowledge work pipeline (content, architecture, research, compliance) can be modelled as a sequence of roles operating on a shared repository.
Live dogfood: forgeyard — yeet2 managing its own development on 10.42.10.101.
- Project registration (clone from URL or attach local path)
- Constitution detection and inspection (VISION.md, SPEC.md, ROADMAP.md + optional files)
- Chat-driven constitution interview (5-question guided setup with LLM synthesis)
- Constitution editor with live markdown preview
- Deterministic and CrewAI-backed mission planning
- 7-stage TDD pipeline: Architect → Implementer → Tester → Coder → QA → Reviewer → (Visual)
- Stage-aware dispatch with role ranking
- Autonomy modes: Manual, Supervised, Autonomous
- Job execution via OpenHands in isolated git worktrees
- Job logs and artifact summaries
- Blocker creation, resolution, and dismissal
- Approval approve/reject workflow
- Pull request creation and merge automation
- Structured QA/reviewer verdict recording
- Stuck job recovery with configurable timeout
- Worker heartbeat and registry
- Markdown rendering across the UI
- Chatroom-style project chat with agent bubbles
- Top navbar + per-project sidebar navigation
- GitHub PAT management via Settings UI
- 21 agent name themes (Star Trek, Star Wars, Stargate, LOTR, Dune, Firefly, and more)
- Character personalities injected into agent stage briefs
- Getting Started guide (in-app + docs site)
- Installation guide (INSTALL.md)
- Docker deployment with health checks for all services
- CI/CD: typecheck, build, Semgrep, Trivy, GHCR multi-arch publishing
- 251+ tests across all services (Vitest + pytest)
- Error feedback and graceful API-down handling across all pages
- OpenRouter model catalog with live pricing
Agent Experience & Visualization
- Mission Control dashboard — real-time grid of all active agents across projects with status (Cursor 2.0 style)
- Live pipeline graph — active node pulses, completed nodes glow green, animated handoffs
- Mid-task chat steering — talk to agents while they work, redirect without canceling (OpenHands/Devin style)
- Chunked progress summaries — narrative updates during execution, not just raw logs
- Spatial agent visualization — characters in themed environments: Star Trek bridge, LOTR war table, The Office floor (Smallville style)
- Execution trace timeline — step-by-step visual replay of agent actions (LangGraph style)
GitHub Integration
- GitHub Projects + Issues kanban — missions as project boards, tasks as issues, stage transitions move cards
- Webhook support — GitHub webhook receiver for two-way sync
Generic Pipeline Platform
- Pluggable execution adapters — passthrough, document, research, shell beyond code-only
- Visual flow editor — n8n/NiFi-style node canvas with drag-and-drop, branching, loops with exit conditions
- Pipeline templates — pre-built pipelines for content, architecture, marketing, legal, data, research
- Chat-driven pipeline design — interview that builds both constitution and pipeline
- pipeline.yml — config-as-code pipeline definition
Extensibility
- Agent Skills / plugins — SKILL.md standard for extending agent capabilities (web search, screenshots, API calls)
- External triggers — start work from GitHub Issues, Slack, webhooks, CLI
- Agent verification artifacts — screenshots, test output, demo recordings to prove work
Observability
- Langfuse integration — self-hosted LLM observability with traces, cost dashboards, quality scoring
- Audit log — searchable history of every autonomy decision
- Cost tracking — per-project and per-role LLM spend from OpenRouter + Langfuse
Operator Tools
- Custom roles — create, rename, reorder roles beyond the defaults
- Role editor UI — drag-to-reorder, per-role model/adapter/character config
- Improved constitution interview — follow-up questions, generate all 6 files
- Worker pool — multiple executor instances with load balancing
- Database migrations —
prisma migrate deployinstead ofprisma db push
- Multi-tenant / multi-user auth
- Nomad-backed distributed execution
- Container-level sandboxing
- Real-time websocket updates
- Plugin system for custom adapters
apps/control SvelteKit web UI
apps/api Fastify REST API + autonomy loop
apps/brain Python planning service (CrewAI + deterministic)
apps/executor Python execution adapter (OpenHands)
packages/db Prisma schema (PostgreSQL)
packages/domain Shared types, model defaults, agent themes
packages/constitution Constitution file discovery and parsing
infra/docker Dockerfiles and deployment scripts
docs/ Product specs, architecture, operations
git clone https://github.com/wan0net/yeet2.git && cd yeet2
cp .env.example .env
# Edit .env: set YEET2_CONTROL_ORIGIN, LLM_API_KEY, LLM_MODEL, LLM_BASE_URL
# Build-on-host deployment:
docker compose --env-file .env -f docker-compose.deploy.yml up -d --build
# Or pull pre-built images:
docker compose --env-file .env -f docker-compose.release.yml pull
docker compose --env-file .env -f docker-compose.release.yml up -dOpen http://localhost:3000 → follow the Getting Started guide.
See docs/INSTALL.md for full setup instructions.
| Doc | Purpose |
|---|---|
| INSTALL.md | Step-by-step deployment guide |
| GETTING_STARTED.md | First project walkthrough |
| VISION.md | Project purpose and principles |
| SPEC.md | Technical specification |
| PRODUCT_SPEC.md | MVP product definition |
| ROADMAP.md | Milestone breakdown |
| BETA_SPEC.md | Beta release specification |
| ARCHITECTURE.md | Service map and deployment |
| DATA_FLOWS.md | Sequence diagrams for all flows |
| OPERATIONS.md | Operator runtime guide |
| DEVELOPMENT.md | Local development setup |
| CI_CD.md | GitHub Actions and GHCR publishing |
| DECISIONS.md | Implementation decisions log |
BSD-3-Clause