AgentGit

AgentGit is a coding tool that saves share and resume AI coding agent sessions across tools. AgentGit saves AI agent sessions from tools like Claude Code and Codex so they can be shared as read-only links, handed off to others to continue, or controlled remotely, with a public hub of example sessions.

AgentGit is a web service for preserving the full record of an AI agent session rather than only the files the agent produced. Instructions, reasoning, failed attempts and decisions are stored together, so a task can be picked up later by another person or another agent without starting from scratch. The site frames this as the difference between keeping the result and keeping the process. Sessions can be imported from several agent runtimes: Claude Code, Codex, OpenCode, OpenClaw, Hermes and WorkBuddy. In practice the workflow is driven from inside the agent itself. The homepage supplies ready-made prompts that a user pastes into their agent, which then follows AgentGit's documentation to hand off a session for someone else to continue, or to generate a secure read-only link for viewing. No separate client is described for these two flows. A second capability is remote control and collaboration. The agent running on a user's own machine can be opened from a phone or another computer to start, continue and follow work. An invite link lets other people join a workspace, send prompts, watch progress and approve actions. Setup again starts with a pasted prompt that points the agent at the remote control documentation. An Agent Hub lets visitors explore public sessions from other users, with example pages such as a paper-to-code project, a Game of Life rebuild and a table tooling demo. This makes the product part code-review aid, part learning library: readers can see how someone else actually reached a solution, not just the final repository. AgentGit suits developers who work with coding agents and need to hand off work to teammates, document how a result was reached, or move between different agent tools. Compared with sharing a git repository or pasting transcripts into chat, it targets the missing layer of agent context. Pricing, account requirements and privacy controls beyond the read-only link are not described on the homepage, so evaluating those requires the docs or a trial.