
1 / 3
【Overview】 AgentGit is a version control layer for AI coding agent sessions. It takes the conversation you have with a coding agent — Claude Code, Codex, OpenCode, OpenClaw, Hermes or WorkBuddy — and records it in a git-based Agent repository, where every turn becomes a commit and every session becomes a branch. Instead of losing a long debugging run when a terminal closes or a machine changes, the work stays saved as history you can return to, inspect, and pass on. The problem it targets is continuity. Agent sessions are usually trapped in one window on one device, so stopping means starting over, and showing someone what happened means pasting fragments of a transcript. AgentGit turns the session itself into the artifact: it can be resumed from the exact turn where it stopped, handed to another person or another agent, or opened through a read-only link for review without giving away write access. It is built for people who work with command-running coding agents — individual developers who move between agents and devices, teammates who need to pick up each other's work, and anyone who wants to watch how others solve problems through the Public Agent Hub. A chat-only assistant with no command execution is outside its scope, and so is anyone looking for an OS-level sandbox for arbitrary shell commands. 【Key features】 • Session save and version control: Records agent conversation turns as commits on session branches inside a git-based Agent repository. • Session handoff: Passes a saved session to another person or agent so they can continue from the point where it stopped, via a prompt or a link. • Read-only share links: Produces a secure link that lets someone view a session without any write access. • Repository invite links: Grants access to a private repository, and optionally to a branch's session, without needing the person's username. • Read-only share snapshot: Publishes a separate read-only copy of a conversation with optional view limit, expiry date, and passphrase. • Remote Control (Multiplayer Mode): Runs a live agent on your computer from a desktop or mobile browser while invited people send prompts and answer approvals. • File-less setup via agent prompt: A single prompt sent to a coding agent installs AgentGit, saves the conversation privately, and returns a session link plus an invite link. • Public Agent Hub: Lets you browse public agent sessions and see the instructions, attempts, tool calls, results, and decisions behind them. 【Use cases】 1. Save an agent session and share it with someone who needs to see how a problem was approached. 2. Hand off a session so a teammate or another agent continues from the exact turn where you stopped. 3. Create a secure, read-only link when you want a session reviewed but not modified. 4. Drive your agent remotely from your phone or another computer while it runs on your machine. 5. Invite teammates to collaborate on agent work and answer approvals during a live run. 【Pros and cons】 Pros: - Treating each session as a branch makes agent work auditable in a way plain chat transcripts are not, which suits developers who need to explain or revisit decisions. - Handoff and invite links remove the usual username exchange and copy-paste ceremony, so teams can split work across people and devices without losing context. - Read-only snapshots with expiry and passphrase give a measured way to show work to reviewers or outsiders without opening a repository. Cons: - Published history is append-only, so there is no rebase or force-push to clean up a branch once it is out. - Cross-runtime handoff may be lossy, and AgentGit states the losses before converting a session. - Only repository owners can create invite links, and a repository can hold up to 100 active links. - A share link is a snapshot for reading; ongoing updates have to live in a published repository. - Remote Control needs a running daemon (agitd) and at least one supported runtime such as Claude Code, Codex, or OpenCode, and a folder binding is not an OS sandbox. 【FAQ】 Q: Which coding agents does AgentGit work with? A: It versions sessions from Claude Code, Codex, OpenCode, OpenClaw, Hermes and WorkBuddy. Remote Control requires at least one of these supported runtimes running alongside the agitd daemon. Q: Can I let someone read a session without giving them write access? A: Yes. Read-only share links and read-only share snapshots both present a session for viewing only, so the recipient cannot modify the repository. Snapshots can additionally carry a view limit, an expiry date, and a passphrase. Q: Is the history of a published session editable? A: No. Published history is append-only, so there is no rebase or force-push once it is out. Cross-runtime handoff also may be lossy, and AgentGit states the losses before converting a session.
WPS AI is an AI assistant built into WPS Office that helps users write documents, generate presentation outlines and…
Baidu Comate is an AI coding assistant built by Baidu on its ERNIE model. It goes beyond simple autocomplete: it…
UniConverter is an AI media conversion and enhancement tool for video, audio, and images. It combines everyday…
Added
10/19/2026
Views
841
Clicks
6
6 similar tools to consider instead of AgentGit