Paca-AI/paca

Paca – AI‑native, self‑hosted project management

What it is – Paca is an open‑source, Docker‑or‑Kubernetes‑deployable project‑management platform that treats AI agents as first‑class members of a Scrum team. Humans and agents share the same Scrumban board, can be assigned to sprints, pick up tasks, co‑author BDD specs and design docs, and interact through an in‑app chat.

Why it matters – Most existing tools (Jira, ClickUp, Monday, Trello) add AI only as a peripheral chatbot. Paca embeds AI directly into the workflow, letting agents act on tasks, run automations, and update the workspace in real time. The core is lightweight; everything else (custom workflows, UI widgets, AI‑specific logic) is added via a sandboxed plugin system.

Core concepts

• MCP Server – a Model Context Protocol server (@paca-ai/paca-mcp) that exposes the workspace (projects, tasks, docs, members, etc.) to any MCP‑compatible agent (Claude, Gemini, custom agents). No custom API work is needed.
• Plugins – backend plugins are compiled to WebAssembly (WASM) and run in a capability‑based sandbox; frontend plugins are ordinary JS bundles. Plugins can add routes, data models, UI pages, or new automation nodes.
• P‑A‑C‑A cycle – the platform enforces the classic Plan → Act → Check → Adapt loop, with AI agents participating at every stage.
• Role‑based IAM – JSON policies modelled after AWS IAM control access to any resource (project, task, sprint, doc, environment, agent, etc.).
• Static environments – long‑lived containers that persist files and processes across conversations, reachable via an in‑browser terminal, SSH, or port‑forwarding.
• Event‑driven automation – visual n8n‑style graph builder (trigger → condition → action) with built‑in nodes for scheduling, webhooks, AI dispatch, and API calls; plugins can contribute additional nodes.

Key features

• Unified Scrumban board where humans and AI agents appear side‑by‑side.
• In‑app AI chat for natural‑language task/epic creation, documentation, and sprint planning.
• BDD (Gherkin) editor co‑authored by agents.
• System Design Documents (SDD) that stay in sync with the board.
• WASM‑sandboxed plugin marketplace (install with one click).
• Real‑time updates via Socket.IO.
• OpenHands‑powered agents running in isolated containers.
• Browser extension to annotate preview pages and turn comments into tasks.
• Role‑based IAM with JSON policies and condition support.
• Jev AI integration for auto‑routing, auto‑filling, and auto‑assigning tasks.
• Support for any Agent Client Protocol (ACP) CLI (Claude Code, Gemini CLI, etc.) via a local bridge.

Installation

1. Docker Compose (single‑node) – run the provided one‑liner:

   curl -fsSL https://github.com/Paca-AI/paca/releases/latest/download/install.sh | bash
   

   The script interactively asks for secrets (JWT, DB passwords, encryption key) and spins up the stack (Postgres, Redis, Caddy, Paca services). For CI/automation set PACA_YES=1 and supply all values via environment variables.
2. Kubernetes – install the official Helm chart:

   kubectl create namespace paca
   helm install paca oci://ghcr.io/paca-ai/charts/paca -n paca -f my-values.yaml
   

   The chart expects publicUrl and secret values (jwtSecret, adminPassword, encryptionKey, …) – no defaults are provided.
3. Upgrade – download the upgrade.sh script from the latest release and run it in the directory containing your docker-compose.yml (or use Helm upgrade).

Extending Paca

• Backend plugins – write any language that compiles to WASM (Go, Rust, AssemblyScript). Place the .wasm file under plugins/backend/ and install via the UI Marketplace or scripts/install-local-plugin.sh.
• Frontend plugins – standard JS/TS bundles under plugins/frontend/ that can add pages, board widgets, or custom settings panels.
• MCP clients – any agent that implements the Model Context Protocol can be pointed at the MCP server (http://<host>:8080) and will gain structured read/write access to the workspace.

Documentation & community

• Architecture overview: docs/architecture/overview.md
• Role & policy guide: docs/guides/roles-and-policies.md
• Environment management: docs/ai-agent/environment-management.md
• Plugin development: see the plugins/ folder and Marketplace UI.
• Contributions are welcomed via the CONTRIBUTING.md guide; the project is Apache‑2.0 licensed.

Bottom line – Paca is a fully open‑source, self‑hosted alternative to commercial PM tools that puts AI agents on equal footing with human teammates, offering a plugin‑first, IAM‑secured, and Kubernetes‑ready platform for modern, AI‑augmented software teams.