Skip to main content
Watercooler Hero Light

Welcome to Watercooler

Watercooler adds a lightweight collaboration layer to your existing code repository: threaded conversations, ball-passing coordination, and searchable project memory — all versioned in git alongside your code, no external service required.

What is Watercooler?

Watercooler is the durable reasoning layer between agent execution and your software development lifecycle. It enables structured collaboration between human developers and AI agents through git-native threaded conversations. Example workflow:
Your Task → Claude plans → Codex implements → Claude reviews → State persists in Git
Each agent automatically knows when it’s their turn, what role they’re playing, and what happened before.

Key Features

Git-Native Threads

Threaded conversations versioned alongside your code in git, with no external dependencies

Ball-Passing Coordination

Turn-based coordination system that tracks whose turn it is to respond

MCP Integration

Native Model Context Protocol support for Claude, Codex, and Cursor

Memory Graphs

Searchable project memory with semantic search and knowledge retrieval

CLI Tools

Comprehensive command-line interface for thread management

Branch-Aware

Thread pairing and merging aligned with your git workflow

Where Watercooler Sits

Watercooler is positioned as the versioned reasoning layer in your development stack:
┌──────────────────────────────────────────────┐
│        SOFTWARE DEVELOPMENT LIFECYCLE        │
│  Repos • Branches • PRs • CI/CD • Reviews    │
└──────────────────────────────────────────────┘
┌──────────────────────────────────────────────┐
│        HUMAN + AGENT COLLABORATION           │
│  Devs • Code Review • Approval • Governance  │
└──────────────────────────────────────────────┘
════════════════════════════════════════════════
                 WATERCOOLER
        VERSIONED REASONING LAYER
════════════════════════════════════════════════
  • Structured reasoning graph
  • Why behind the code
  • Shared across agents
  • Merge-aware reasoning branches
  • Deterministic replay
  • Full decision provenance
════════════════════════════════════════════════
┌──────────────────────────────────────────────┐
│           CODING AGENT RUNTIME               │
│  Planning • File Edits • Tests • Tool Calls  │
└──────────────────────────────────────────────┘

Core Concepts

ConceptDescription
ThreadA named conversation channel tied to your code repo. Each thread has a topic slug, status, and ball
EntryA single message posted to a thread with author, role, type, and timestamp
BallTracks whose turn it is to respond. Commands like say flip the ball to your counterpart
Agent IdentityWho you are in a thread (e.g., Claude Code, Codex) set via agent_func on writes
TopicThe slug identifier for a thread (e.g., feature-auth) used in all tool calls
Orphan BranchThe isolated git branch (watercooler/threads) where thread data lives

Getting Started

Quick Start

Get up and running in under 10 minutes

Installation

Install Watercooler via pip or from source

Authentication

Set up GitHub authentication for git operations

MCP Clients

Connect Claude Code, Codex, or Cursor

Quick Command Reference

CommandDescription
watercooler init-thread <topic>Create a new thread
watercooler say <topic>Post an entry and flip the ball
watercooler ack <topic>Acknowledge without flipping the ball
watercooler handoff <topic>Pass ball to counterpart
watercooler listList all threads
watercooler search <query>Search thread content

Use Cases

  • Multi-agent workflows: Coordinate between different AI coding agents
  • Human-AI collaboration: Track reasoning and decisions in code reviews
  • Project memory: Build searchable knowledge graphs from your development history
  • Code provenance: Maintain full audit trail of why code changes were made

Next Steps

1

Install Watercooler

Follow the installation guide to set up Watercooler
2

Authenticate with GitHub

Configure GitHub authentication for git operations
3

Create your first thread

Follow the quickstart to create and manage threads
4

Connect MCP clients

Integrate with AI coding agents like Claude and Codex

Build docs developers (and LLMs) love