Paca: An Open-Source, AI-Native Alternative to Jira

AI Agents as First-Class Scrum Teammates

Paca is a self-hosted project management platform that integrates AI agents directly into the Scrum process rather than treating them as peripheral chatbots. Unlike traditional tools where AI is an add-on, Paca allows AI agents to be assigned to sprints, pick up tasks from the backlog, and update statuses in real time alongside human teammates.

AI agents in Paca contribute to the development lifecycle by:

  • Collaborating on BDD specs: Helping Product Owners and Business Analysts write Gherkin scenarios.
  • Contributing to System Design Documents (SDD): Maintaining visible architecture documentation to keep the team grounded.
  • Executing tasks: Pulling work from the board and providing updates based on the project's current state.

The P-A-C-A Collaboration Cycle

Paca organizes team collaboration around a four-phase cycle designed to mirror both the scientific method and the Scrum framework:

Phase Activity
Plan Product Owners, BAs, and AI agents collaboratively refine the backlog and co-author BDD scenarios and SDD designs.
Act Humans and AI agents pull tasks from the live sprint board, execute them, and post updates.
Check QA agents perform automated verification while humans review AI-generated output.
Adapt The team conducts retrospectives using sprint data to inform the next cycle.

Technical Architecture and Extensibility

Paca is built with a lightweight core and a highly modular architecture to avoid the feature bloat common in enterprise project management software.

System Stack

  • Frontend: React, TanStack Start, and shadcn/ui.
  • Core API: Go with the Gin framework.
  • Real-time Layer: Node.js and Socket.IO for event fan-out.
  • AI Orchestration: Python, FastAPI, and the OpenHands SDK, where agents execute in isolated sandbox containers.
  • Infrastructure: PostgreSQL for persistence and Valkey for caching and async event streams.

Customization via WASM Plugins

Everything in Paca is configurable via project-level files. For deeper extensions, Paca uses a plugin system where backend plugins are compiled to WebAssembly (WASM). This allows developers to write plugins in Go, Rust, or AssemblyScript that run in a sandboxed environment with a capability-based permission model, ensuring plugins only access declared host functions.

AI Integration: MCP and Claude Code

Paca provides deep integration for external AI tools to interact with the project data layer without requiring custom API wiring.

Model Context Protocol (MCP) Server

Paca includes an MCP server (@paca-ai/paca-mcp) that allows compatible AI agents (such as Claude Desktop) to perform structured operations. Available tools include:

  • Project Management: Creating, updating, and listing projects and members.
  • Task & Sprint Control: Managing tasks, sprints, and custom fields.
  • Documentation: Handling System Design Documents and attachments.
  • Activity Tracking: Adding and updating task comments and activities.

Claude Code Skills

For developers using Claude Code, Paca offers a set of slash commands (e.g., /paca-epic, /paca-breakdown, /paca-sprint) that allow users to manage the workspace directly from their editor. These skills read Paca documentation first to ensure the AI is contextually aware of the project before executing commands.

Version 0.4.0 Updates

The v0.4.0 release introduced two primary features to improve project planning and auditing:

  • In-app AI Chat: Users can now chat with AI agents at the project level to create or update epics, stories, and tasks using plain English.
  • Activity Diff & Revert: The activity pane now displays a visual before/after diff for every field change, allowing users to revert changes with a single click.

Community Perspectives and Analysis

Discussion among users highlights a shift toward "vibecoding" and the need for lightweight, agent-centric tracking over heavy enterprise tools.

"I'm using GitHub issues and GitHub Projects with gh cli and I find it works well, though what I really like about this is your project level chat." — @hmokiguess

"Feels like it's geared toward actually enabling the 'dark factory,' which is pretty difficult with enterprisey, seat based SaaS like GitHub and Jira." — @Jgrubb

Some users expressed caution regarding the adoption of new project management tools, noting that workflows are often highly personalized. Others questioned the arbitration process when humans and AI agents disagree on task priority, and the security implications of placing proprietary data in an AI-accessible environment.

Sources