Version control for AI coding sessions

Don’t lose the why behind your code.

Entrpy versions AI coding context with the code it belongs to, so developers and small teams can resume work, hand tasks off, and keep decisions tied to the right repo state.

Early product phase · starting with Codex CLI, local MCP access, and Git-based sharing.

Request early access See the workflow

repo / agent-session

# work with your coding agent
$ codex
...

# Entrpy stores the useful context
$ entrpy snapshot
✓ session captured
✓ decisions linked to commit 8f31a2c
✓ local context ready for handoff

# another developer continues later
$ entrpy context
→ architecture decisions
→ attempted approaches
→ next recommended step

Session historyKeep useful agent work instead of losing it in chat windows.

Code-awareContext stays associated with the branch or commit it came from.

Git-nativeShare selected history through the repo workflow teams already use.

Built for developersIndividual builders and 2–10 person software teams using AI coding agents.

CLI firstStart inside the workflow instead of adding another dashboard to maintain.

Focused MVPResume and hand off AI-assisted work before expanding into a larger collaboration layer.

The problem

AI can write the next line. It still forgets how you got here.

Teams increasingly move between sessions, tools, branches and people. The code survives. The reasoning, rejected approaches and task context often do not.

Sessions disappear

Important implementation context stays trapped in a previous agent conversation or gets manually copied into notes.

Handoffs restart the work

A teammate sees the current code, but not the sequence of decisions that made that code the current state.

Context drifts from code

Old reasoning can be misleading when it is no longer tied to the branch, commit or implementation it described.

How it works

A lightweight history for AI-assisted development.

Entrpy is being built around the minimum loop needed to make previous AI work reusable: capture, version, continue.

Capture

Record the useful context

Save session history, explicit decisions and implementation rationale automatically as you work.

Version

Tie it to code state

Associate records with the relevant repo state so outdated decisions are not presented as current truth.

Continue

Resume or hand off

Use local MCP access and Git-based sharing to continue the work from prior context instead of rebuilding it manually.

Product direction

Git for code. Entrpy for the agent context around it.

The first goal is not a giant platform. It is proving that versioned agent context can reduce repeated work and make handoffs measurably better.

What we are building first

A focused CLI MVP for real development workflows, starting with Codex CLI and a local MCP connection.

Automatic session records after project setup
Decision and rationale history linked to code state
Local retrieval of relevant past context
Selected history shared through Git commit / push / pull
Pilot testing with individual resume and teammate handoff scenarios

What success looks like

Developers should be able to return to a task on another day or inherit a teammate’s work without asking for a full verbal reconstruction of what happened before.

Less repeated prompting and rediscovery
Faster continuation across sessions
Cleaner handoffs between developers
Context that respects the current branch and commit

Early access

Help us test whether AI work can become reusable team history.

We are looking for developers and small software teams that use AI coding agents repeatedly and feel the cost of lost context, repeated prompting or difficult task handoffs.