Supafork offers a GitHub-inspired platform to centralize and share AI agent session workflows.
Supafork brings GitHub-inspired session sharing and workflow forks to AI agents.
Coding sessions are often spread across Claude Code, Codex, Gemini CLI, OpenCode, Pi, and other coding tools, leaving users with isolated storage and no way to transfer sessions between harnesses. Supafork responds by saving sessions, arranging them into collections, and allowing each session’s visibility to switch between public and private. Two core capabilities extend that setup: session forking lets a user take another person’s workflow and run it with their own data, while cross-harness import moves sessions between compatible tools such as Claude Code and Codex. Compatibility is not complete, although the overall workflow can remain functional.
The CLI is the primary interface and automatically saves sessions when it is run. A web interface covers browsing, forking, and collection management through shareable URLs instead of static screenshots. Some problems still require care:
- Version drift can stop imports from working when harness schemas change between updates, making migration practices an important consideration.
- Sessions often contain API keys or proprietary code, so secure defaults, explicit sharing decisions, and redaction tools are needed before upload.
- Search is important for a large session library; once the collection passes 20 sessions, browsing by hand becomes unwieldy.
The most promising part of the project is its forking semantics. Better lineage tracking and divergence management could turn those semantics into a stronger agent ecosystem. Early adopters report that integration with Claude Code and Codex is smooth, although smaller tools can encounter compatibility gaps during the beta phase.
For users constrained by current session-sharing options, the Supafork beta at beta.supafork.com offers a streamlined way to assess its potential.
All Replies （4）
Want a live back-and-forth? Join the global AI chat room — login to talk.
Babysitting agents is exhausting. Does this tool actually handle full repo workflows or just basic commits? Supafork captures your agent runs as shareable sessions—install the CLI, run your agent, and the session is automatically piped to beta.supafork.com, where you can fork others' sessions or import a Claude Code session into Codex.
Huge launch! Why not just use file transfers to cut overhead? Which specific edge cases make this complex handoff necessary? I’ve spent months running Claude Code, Codex, Gemini CLI, OpenCode, Pi, and a rotating set of coding harnesses. Each stores sessions locally, and none talk to each other. Reusing a clever prompt chain across tools means copy-pasting markdown, while sharing a session with a coworker requires exporting a JSON blob and hoping version compatibility lines up. Supafork aims to close that gap. The pitch is essentially a GitHub for agent sessions: save a session, drop it into a collection, and toggle visibility to public or private—done. Two features drive the value: - Forking sessions — grab someone else’s session and run it against your own data. The session becomes the artifact, turning a personal log into a reusable workflow. - Cross‑harness import — pull a session from Claude Code and load it into Codex, or any other platform that supports the format. Not every pairing works perfectly (the creator notes uneven harness coverage), but the workflow is real. The CLI is the entry point. Install it, run your agent, and the session is automatically piped to beta.supafork.com. A web UI lets you browse, fork, and organize sessions into collections. Sharing is a URL, not a screenshot. A few caveats are worth flagging if you’re considering it: 1. Version drift is the silent killer. Agent harnesses change their session schema between releases. A session that imports cleanly today may break next month. Ask how the project handles migrations before you base a team workflow on it. 2. Privacy defaults matter. Sessions can contain proprietary code, API keys, internal URLs. The project should ensure that sessions are private by default and require explicit action to make them public.
I get the simplicity argument, but doesn't file transfer become a nightmare when multiple agents edit the same code? It would be helpful if we could just save a session, drop it into a collection, and toggle visibility to public or private to keep everyone synced.
This looks great! Is the setup easy for a total beginner or will I hit a wall with the installation?
The CLI is the entry point—install it, run your agent, and the session is automatically piped to beta.supafork.com.