Ajapopaja Build

Welcome to Ajapopaja Build – an automated task‑pipeline manager for Coding AI agents, with a human‑supervised management layer. It helps you break work into tasks, let agents design, implement, and review the changes, and approve the results before they land.

---

What is it?

A pipeline of tasks that move through a clear lifecycle (created → scheduled → in progress → implemented). A web dashboard gives you full control, while a set of autonomous agents do the heavy lifting. The project is a monorepo:

• backend/ – a uv workspace (core models, api FastAPI server, ajapopaja_mcp MCP server).
• frontend/ – a Vite + TypeScript Single Page Application (Tailwind CSS).
• design/ – architecture and bot design documents.
• INSTALL.md – full production installation guide (Docker + systemd).

---

Autonomous agents

Agents work on real code inside isolated sandboxes and report back through the UI:

• CoderBot 🧑‍💻 – the end‑to‑end coder. It implements tasks described in design documents by driving the Pi coding agent in headless mode, then submits a pull request for your approval. Accept it (optionally auto‑committing the change) or reject it – all in the dashboard.
• ArchitectureBot 📐 – drafts a design document for a task with one click.
• ReviewBot 🔍 – performs an automated technical review of an implemented task (spec + design doc + git diff) and stores the review in the task.
• DocBot 📝 – creates design documents, code comments, and documentation.
• BotManager ⚙️ – queues and serializes every bot so only one LLM session runs at a time.

Key features

• Pull‑request workflow – every CoderBot change becomes a PR that is reviewed and committed (or rejected) in the UI.
• Human‑in‑the‑loop design review – tasks can require a design document that you approve before implementation starts.
• MCP + Pi integration – AI agents connect over the Model Context Protocol; a pi skill and mcp_client.py CLI make task search/management scriptable.
• Real‑time UI – WebSocket‑driven dashboard with a git‑status indicator, multi‑layout columns, persisted state, and a streaming logs viewer.
• Secure by default – JWT‑based authentication across the API, sandboxed workspaces with path‑traversal protection, and Markdown sanitization.

---

Installation

Production (Docker + Linux systemd service): follow the complete guide in INSTALL.md – it covers building the Docker image, configuring environment variables, and installing/applying the systemd service.

Quick local dev setup:

# Backend (Python 3.11+, uv, MongoDB, Ollama)
cd backend && uv sync

# Frontend
cd frontend && npm install

# Run them
cd backend  && uv run --package api uvicorn api.main:app --reload   # http://localhost:8000
cd frontend && npm run dev                                          # http://localhost:5173

See INSTALL.md and CONTRIBUTING.md for environment variables and development guidelines.

---

Usage workflow

1. In the dashboard, create a Pipeline and point it at a local workspace.
2. Create Tasks – give each a technical spec and choose whether it needs a design document (and which agent should handle it).
3. Let the agent pick up the task: CoderBot implements it, ArchBot drafts a design, ReviewBot reviews it.
4. Review the results (design docs, diffs, PRs) in the UI and accept or reject them.

---

Testing

# Backend
cd backend && uv run pytest

# Frontend
cd frontend && CI=true npm run test

---

Release history

See RELEASENOTES.md. The current release is 0.3.0.

---

License

Licensed under the Apache License 2.0. See LICENSE and LICENSE_COMPLIANCE.md for dependency licensing details.

Note: this repository has a pre-push hook that blocks unintentional pushes. To push to GitHub manually, set ALLOW_PUSH=true git push.