AI News HubLIVE
站内改写2 min read

Mach: Version Control for AI Agents

Mach is an open-source tool that provides Git-like version control for AI agent sessions, enabling push, pull, and clone operations, model switching mid-task, and team collaboration. It cryptographically verifies every agent action, addressing the opacity and isolation of current agent sessions.

SourceHacker News AIAuthor: harsh020

Collaboration Tool for AI

Version Control for AI Agents

Push, pull, and clone agent sessions. Switch models mid-task and collaborate on AI workflows with your entire team.

Try mach (Open Source)Enterprise WaitlistLogin

mach clone origin/session_29a

Branch:feature/auth-gen

Fetching remote state

Swapping model gemini 1.5

Resuming agent task

stdout

$mach clone origin/session_29a

Cloning into 'working-dir'...

remote: Enumerating agent states: 45, done.

Receiving state: 100% (45/45), done.

✓ Switched to branch 'gemini-handoff'

The Problem

Why now?

AI agent sessions are isolated, opaque black boxes. When an agent hits a wall, you can't easily hand off its context. When it makes a mistake, enterprises have no way to audit its reasoning. Traditional version control tracks human code, not autonomous agent intent.

Treat agent memory like a git repository.

Cryptographically verify every agent action.

Collaborative Agent WorkflowDistributed

Pushed session: claude_ui_build

User A • 10:41am

Pulled session locally

User B • 10:45am

Resumed task with: gemini-1.5-pro

Gemini • 10:46am

The Protocol

How Mach Works

Mach establishes a standard protocol for serializing, versioning, and verifying agent memory, allowing seamless operations across any model or platform.

  1. Wrap & Record

Agents run in a Mach-wrapped environment that deterministically records context, memory, and actions as Git-like commits.

  1. Cryptographic Proof

Every agent step is hashed and linked, creating an immutable, verifiable ledger of autonomous intent and execution.

  1. Push & Collaborate

Push sessions to the remote registry. Team members can pull, branch, and swap to different local or remote models seamlessly.

Endless Possibilities

Unlock entirely new workflows when agent memory is decoupled from the execution engine.

Model Swapping

Start drafting a document with Claude's long context. Hit a coding blocker? Commit the session, pull it, and resume the exact same context window using Gemini or a specialized fine-tuned model.

CL

GM

Llama

Multi-Agent Orchestration

Allow a specialized researching agent to generate a session tree, and let independent coder agents branch from specific commits to build implementations in parallel.

Time-Travel Debugging

Agent hallucinated and deleted critical code? Checkout the session state right before the error and steer it in the right direction.

Visual Agent Session Trees

Every session is a ledger of sequential steps: inputs, reasoning traces, tool executions, and outputs. Branch off at any step to explore different outcomes, test parallel prompts, or swap models mid-task.

Platform Overview

mach OSS

Local-first version control for agent memory, context, and tasks.

Push, pull, clone active sessions

Handoff tasks between AI models

Standardize states across platforms

cognatoai/mach

Developer Cloud

Hosted registries to store, share, and track agent sessions effortlessly.

Cloud-hosted session registries

Seamless sharing of agent workflows

Web-based session history viewer

Integration with dev pipelines

Join the Waitlist

Enterprise

Full auditability, compliance, and access control for your multi-agent workforce.

Immutable agent audit logs

Role-based access & SSO

SOC 2, EU AI Act, GDPR ready

Visual node-graph & risk engine

Request Access

For Developers oss cli|For Teams Developer Cloud|For Enterprise Compliance Suite

Ready to collaborate with AI agents?

Join the waitlist for early access to the Enterprise platform.