Cognato AI
Waitlist
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 Waitlist
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.

2. Cryptographic Proof

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

3. 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.

{ user_prompt }[ reasoning ]> tool: read_file{ session_merge }[ spawn: test_run ]> tool: bash{ output: failed }[ swap: gemini-1.5 ]> tool: write_file

Platform Overview

mach

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

  • Push, pull, clone active sessions
  • Handoff tasks between AI models
  • Standardize states across platforms
harsh020/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.