You know that sinking feeling. You cd into the project directory, scroll through your terminal history, and realize the conversation you had with your coding agent—the one that cracked that impossible bug, the one with the perfect refactoring plan—is gone. Buried. Lost in the ether of a single tool’s chat log.
I’ve been there. We’ve all been there. And the worst part? It’s not your fault. It’s by design.
Every coding agent—Codex, Claude Code, Pi, OpenCode, Grok—wants to own your context. The more you pour into that conversation, the stickier it gets. The more you need that session, the more the tool has you hooked. This isn’t a bug. It’s the feature they’re not advertising.
Your accumulated context belongs to you or to the tool? That’s the question that determines whether you’re a user or a product.
I built agent-hop to solve my own frustration. It started simple: search across all your coding agents’ chats from one place. But then I realized something. If I can read the session formats, I can write them too. I can take a conversation that started in Codex and pick it up right where it left off in Claude Code. Or in Grok. Or in OpenCode.
So I reverse-engineered the session formats for each of these agents. I built adapters. And now agent-hop lets you do exactly that: resume any session in any agent.
This is more than a convenience hack. This is a statement about ownership.
The next real moat in AI coding tools isn’t the model — it’s session memory. Reverse-engineering session formats commoditizes the agent layer and makes model quality the only remaining differentiator.
Think about it. The big players are spending billions on training better models. But the thing that keeps you locked in isn’t the model’s intelligence—it’s the history you’ve built with it. The context. The shared understanding. The session.
By making that session portable, we flip the power dynamic. Suddenly, your choice of agent isn’t a prison sentence. It’s a preference. You can start with one, switch to another, come back. The conversation belongs to you, not to the tool.
I’ve seen developers juggle three different agents in a single afternoon, each one better at a different task. But they lose the thread every time. They retype the same context. They restart from zero. It’s a massive tax on productivity—and it’s entirely unnecessary.
Agent-hop isn’t just a search tool. It’s a key that unlocks your own data. It’s the first step toward a world where AI agents are interchangeable interfaces, not walled gardens.
The question isn’t which model is best. It’s who owns your memory.
Try it. Break a session out of one agent and drop it into another. See how it feels. Then ask yourself: should any tool hold your history hostage?
FAQ
Q: Is this really a vendor lock-in issue? Aren't sessions just a convenience feature?
A: Sessions are the accumulated context that makes an AI agent useful. The longer you use one tool, the more valuable that context becomes. If you can't move it, you're locked in. It's a subtle but powerful moat.
Q: How does agent-hop actually work? Does it require access to the agent's internal data?
A: Agent-hop reads the local session files that each agent stores on your machine. These are just files—databases, JSON, etc. It reverse-engineers the format and then writes it into the format of another agent. No API access needed, just your own files.
Q: Won't the agents just change their session formats to prevent this?
A: They could, but that's a cat-and-mouse game. The community can update the adapters. More importantly, making formats opaque hurts users. The real solution is an open standard for session portability. Until then, reverse-engineering is the only way.