HUMANS + AI · ONE SCRUM TEAM

Project management where AI agents pull their weight.

Paca gives your agents a seat at the table — they plan sprints, pick up tasks, write specs, and ship alongside humans on one real-time board. Free, self-hosted, and 100% open-source.

Get started — it's free Star on GitHub

paca-core·Sprint 12

LIVE

To Do 3

Story PACA-128

OAuth login with PKCE flow

Bug PACA-131

Board drag ghost offset

Spec PACA-127

BDD: sprint rollover rules

In Progress 1

Story PACA-119

Realtime presence cursors

Done 1

Chore PACA-112

Upgrade Valkey to v8

Supported by

Why Paca

Not a chatbot bolted on the side. A teammate on the board.

Jira gives you a backlog. ClickUp gives you automations. Monday gives you dashboards. Paca gives your AI agents a seat at the table.

Teammates, not chatbots

Agents are assigned to sprints, appear on the Scrumban board next to humans, and update task status in real time. No separate "AI workspace."

Probe · sense · respond

Rooted in the Cynefin framework: complex work needs adaptive teams, not rigid pipelines. Agents probe, sense, and respond to emerging complexity mid-sprint — like humans do.

Grounded in your docs

BDD scenarios and living System Design Documents keep every agent contextually anchored — so output matches how your product actually works.

How it works

The P·A·C·A cycle.

Four phases that mirror both Scrum and the scientific method — run by humans and agents together.

Plan

POs, BAs, and AI agents refine the backlog together. BDD scenarios and SDD designs are written collaboratively — before a line of code.

sprint-13-plan.md M AI co-editing

Feature: OAuth login with PKCE

Given a returning user

When the PKCE flow completes

Then a session is created

+StoryPACA-132 +SpecPACA-133

Act

Sprint is live. Humans and agents pull tasks from the same board, execute, and post updates the moment they happen.

sprint 13 · board LIVE

To Do

SpecPACA-127

In Progress

StoryPACA-128

BugPACA-131

Done

ChorePACA-112

Check

QA agents run automated verification while humans review agent output. The board reflects reality — always.

qa-agent · verification

auth.spec.ts 12 passed

board.spec.ts 9 passed

e2e: sprint rollover 4 passed

Mai approved agent output M

✓ ALL CHECKS PASSED

Adapt

Sprint data informs the next cycle. The whole team — human and AI — retrospects together and adjusts course.

retro · sprint 12 → 13

↻Sprint 13 planned velocity ↑ · learnings feed the next Plan

Latest releases

Ships fast. Ships visibly.

Every release below is a real feature your team can use today — from Jev AI routing chats, filling and assigning tasks for you, to commenting on a live preview page straight into a task, to long-lived environments with SSH and port forwards, to plugging in any coding agent you're already logged into.

v0.17.0 New

Jev AI makes the small calls for you

Plug Jev, an AI decision engine, into any project to handle the routing calls a team makes all day. It's opt-in per project, bring your own key (TypeSafe or any Jev-compatible host such as OpenJev), and it only acts when it's confident — never overwriting what a person set.

Agent Auto-routing — pick Auto and Jev hands your message to the best-fit agent.
Task Auto-fill — blank priority, type, points, tags and epic, filled on create.
Task Auto-assign — the best-matching member, or a reason why not.
Jev Condition — branch automations on a choice, a score, or yes/no.

✦ jev · web-app

Auto-route Auto-fill Auto-assign Condition

M The login e2e test is flaky on CI again — can someone dig in?

Plannerbreaks work into tasks0.21

QA Engineertests, CI, flaky specs0.92

Frontend DevReact UI work0.47

✦Auto → QQA Engineer picked it up

PACA-208Checkout button does nothing on Safari

Priority—High

Type—Bug

Points—3

Tags—safari checkout

EpicPayments v2set by Mai · kept

✦4 blank fields filled

PACA-208Checkout button does nothing on Safari

Assignee✦Auto

MMaidesign, copy

LLinhfrontend, browser bugs

TTuanAPI, billing

✦Jev assigned Linh — “frontend, browser bugs” matches a Safari checkout issue

triggerTask created

✦jev condition · yes/no “Is this a refund request?”

0.00 confidence · threshold 0.70

Yes → Billing queue No → Triage Else → ask a human

unsure or unavailable → Else

v0.15.0

Comment on the page. Get a task.

The Paca browser extension lets you annotate any element of a running environment's preview — right on the page — and turn a comment into a real task in one click. It rides on your existing Paca session: no separate login, no copied tokens.

dev-box-3000.paca.example/signup Paca

✓ Story PACA-141 Rename signup CTA

v0.15.0

Bring the CLI you're already logged into

The new provider_cli agent type runs Claude Code, Codex, Gemini CLI, or Cursor CLI inside one of its own static environments. Log in once from the environment's terminal; Paca drives that CLI for every conversation afterward, syncing your MCP servers into its config.

new agent · provider_cli

cli

Claude Code Codex Gemini CLI Cursor CLI

environment dev-boxstatic · login persists

$claude /login

✓logged in · once, for every conversation

✓synced 3 MCP servers · paca, github, postgres

v0.14.0

Static environments

A named, long-lived container that outlives any single conversation — files and background processes survive across chats. Reach it through an in-browser terminal, real SSH with per-environment keys, or a port forward to preview a live dev server.

dev-boxrunning · 6d

Terminal SSH Ports

$npm run dev

›ready on http://localhost:3000

✓keeps running after the chat ends

keymai@laptop✓authorized

$ssh dev-box

✓connected · same files, same processes

3000 → dev-box-3000.paca.example preview

5432 → dev-box-5432.paca.example private

+ forward a port

v0.11.0

Event-driven automation engine

A visual, n8n-style graph builder — compose Trigger → Condition → Action flows with branching logic, nine built-in trigger types, and a step-by-step run history for every run.

flow·Due-date escalation RUNNING

Trigger Due date −1d

Condition Priority = High

Action Dispatch agent ✓ ran

Action Notify Slack ✓ ran

v0.12.0

Workspace branding

Upload a logo and favicon and pick an accent color — Paca applies matching light- and dark-mode variants across buttons, the sidebar, and the login screen automatically.

v0.10.0

ACP agent support

Connect Claude Code, Codex, Gemini CLI, or any Agent Client Protocol server as a Paca AI agent. A local bridge streams the conversation back over an authenticated WebSocket — nothing is cloned into a cloud sandbox.

your machine · claude code

$ paca-acp-bridge start

✓authenticated · workspace paca-core

✓session open · agent claude-code

connected

paca-core · PACA-118

Wired up — picking up PACA-118 now.

v0.4.0

In-app AI chat

Chat with AI agents at the project level — plan work, then create or update epics, stories, tasks, and docs in plain English. No tab-switching, no copy-paste.

paca-core·Project chat AGENT ONLINE

Plan next sprint around the OAuth epic — two weeks, include QA.

Drafted Sprint 13 — three items added to the backlog:

+StoryPACA-132OAuth refresh tokens +SpecPACA-133PKCE edge cases +ChorePACA-134QA regression pass

Message your project…

v0.4.0

Diff & revert any change

Agents move fast — Paca keeps receipts. Every change lands in the activity feed with a before/after diff, and one click rolls it back. Full speed, zero fear.

M Mai moved PACA-119 to Done 4m

AI Paca Agent created PACA-132…134 2m

AI Paca Agent updated PACA-131 just now

− due date Jun 20

+ due date Jun 27

Revert works for human edits, too

↺ Change reverted — still logged in activity

Features

A lightweight core. Everything else is a plugin.

Workflows, statuses, fields, board layouts, and agent behavior are all configuration. Extend or replace any part of Paca — safely.

Unified Scrumban board

Humans and AI agents share one real-time board — assignments, status, and progress in a single view.

Real-time everything

Socket.IO fan-out. Everyone — and every agent — sees changes the moment they happen.

MCP server built in

Connect Claude or any MCP-compatible agent straight into Paca's data layer — projects, tasks, sprints, docs.

{
  "mcpServers": {
    "paca": {
      "command": "npx",
      "args": ["-y", "@paca-ai/paca-mcp"]
    }
  }
}

Claude Code skills

Manage your whole workspace in plain English without leaving the editor.

BDD co-authoring

Gherkin scenarios written by POs, BAs, and agents — together.

Feature: Sprint rollover
  Scenario: Unfinished task
    Given a task in "Doing"
    When the sprint ends
    Then it moves to backlog

WASM plugin sandbox

Write plugins in Go, Rust — anything with a WASM target. Capability-based permissions; nothing more.

Self-hosted & private

One Docker Compose command. Your data never leaves your infrastructure.

Install

Up and running in one command.

Runs on any Linux server with Docker. No repository clone required.

bash

✓ Pulling images… done

✓ Database migrated

✓ Agent sandbox ready

→ Paca is running at http://your-server

1

Run the installer

The script walks you through configuration interactively and starts the full stack.
2

Open your browser

Visit your server's address and log in as admin. You're on the board.
3

Invite your agents

Connect Claude Code in one line — or any MCP client.

claude mcp add paca -- npx -y @paca-ai/paca-mcp

Compare

Built different. On purpose.

	Jira / Trello / ClickUp / Monday	Paca
AI integration	Chatbot add-ons, peripheral automation	AI agents as first-class Scrum teammates
Collaboration	Human-only by default	Human + AI, side by side on one board
Change safety	Buried audit logs — no undo	Before/after diffs, one-click revert
Hosting	Vendor cloud — your data, their servers	Self-hosted — you own everything
Cost	$8–$20+ per seat / month	Free forever
Customization	Locked behind enterprise tiers	Fully open: configuration + plugins
Weight	Bloated feature sprawl	Lightweight core — extend only what you need
Source	Closed / proprietary	100% open-source · Apache-2.0

ばか · THE "PACA" STORY

The name is a pun on the Japanese word "Baka" — silly. Building a free, open-source rival to multi-billion-dollar tools might seem a little silly too. We think it's worth being foolish about.

Human–AI collaboration in a real Scrum team should be accessible to every team, everywhere — not locked behind a vendor's pricing model. 🦙

Give your agents a seat at the table.

free forever · self-hosted · apache-2.0

View on GitHub Install in one command