Introducing Supafork

Git Was Not Built for the Agentic Era

The volume and frequency of code have increased exponentially. The form factor of coding with AI has completely transformed, and we need a new kind of system where agentic coding and version control work hand in hand.

CLI changed the way we code forever. Going from staring at code for hours to YOLO-ing your way out through a CLI prompt is insane. Our attention to code is going away, but our need to access it when needed is not. Imagine a system where you can access all the information your agent produces.

#The Graveyard of Tokens

Humans are rarely coding these days, and agents are doing the heavy lifting. All of that rich information you give your agent is gold. Whether the agent does a good job with it or not, which heavily depends on the model and the harness, those tokens are sacred.

All the tokens a human generates throughout their use of agents need a home. They need a permanent address. Supafork is that.

#One Fork to Rule Them All

Agentic harnesses have risen, and this is just the beginning. We are at the beginning of making agents actually conscious, and the way to do that is yet to be figured out. But coding harnesses are one of the earliest signals that it is possible.

Your agents should be capable of running anywhere and everywhere, existing independently and autonomously. To solve that, the construct we need to figure out is: what makes an agent?

Filesystem + Metadata + Runtime = Agent

Broadly, runtime is already being solved by large infrastructure players like Daytona, Modal, and others. The filesystem and metadata should be able to exist on any runtime, and the only way to make that happen is to somehow package this data into a file and plug it in anywhere.

Across all the harnesses that run agents, there needs to be a common place. One ring to rule them all.

#Built for the Agent Economy

We believe the agent economy is going to be highly decentralized, and no single player will actually win. It will be extremely volatile, with a lot of room for each player. By 2030, we believe the number of agents will grow to 100B.

While that number seems very steep, we are getting very close to the critical point that will make this take off like never before. For that economy to exist, we need the infrastructure to be prepared. We need the frameworks and protocols for how agents actually exist.

The biggest missing piece in the puzzle is where all this data is going to be stored and how it will be consumed. We believe immutable agent storage is the answer to this.

#Our First Step

Among all the things that we anticipate needing to solve, we have a clear winner for what to solve right now: everything an agent does.

We are building a common application that talks to all the harnesses and acts as the centralized storage layer for metadata. Be it your agent data, sessions, prompts, or harness configurations, we make sure you have access to this data and make your agents omnipresent.

#Build with Supafork

Supafork is CLI-installable. It is a one-time setup to get started. You run supafork init or sf init to get started on your repo and make it an Agentic Repo. You can then use sf session and sf session <id> to access session context from any harness and move between harnesses.

Today, we support OpenCode, Codex, Claude Code, Pi, Gemini, and Cursor, and we are adding support for more harnesses.

You can access these sessions and share them with your friends and colleagues. They can fork your session and continue where you left off. We are laying the path toward agent Multiplayer.

Here goes Supafork’s CLI:

▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄█▀▀▀█▄▄▄▄▄▄▄▄█▀█▄▄▄
█ ▄▄▄█ ██ █ ▄▄ █▄▄▄ █▀ ▀██ ▄▄ █ ▄▄█ █▀ █
█▄▄▄ █ ██ █ ██ █ ▄▄ ██ ███ ██ █ ███ ▄ ▀█
█▄▄▄▄█▄▄▄▄█ ▄▄▄█▄▄▄▄██▄███▄▄▄▄█▄███▄██▄█
          ▀▀▀

supafork v0.1.0

Usage:
  supafork [command]

Commands:
  login    Log in with Google (one session for the whole CLI)
  init     Mark the current directory as a shareable supafork repo
  session  Browse native coding-agent sessions for the current project
              list sessions from supported harnesses, or inspect one by id
  agent    Manage coding-agent definitions across harnesses
              list agents from every harness, or add/delete them in the
              harness's own store (claude-code, codex, opencode)

Flags:
  -h, --help      Show help
  -v, --version   Show version

Run 'supafork help <command>' for command details.

Follow @supafork:
  Website    <https://supafork.com>
  Twitter    <https://x.com/supafork>
  LinkedIn   <https://linkedin.com/company/supafork>