I use more than one coding agent. Claude Code, Codex, Grok, sometimes Antigravity. The problem was never which one is best. It was that every time I switched, I lost the whole conversation. I would copy a summary by hand, hunt for a session id, and start the new agent cold.
So I built a small open source tool to fix exactly that, and this is how it works, plus the part that broke while building it.
The idea: move the delta, not the transcript
The lazy way to move context between two agents is to copy the entire conversation. It is also the wrong way. A transcript is large, it repeats what the target already knows, and it buries the two sentences that actually matter.
context-bridge keeps a small knownBy matrix instead: for every pair of agents, how far into each agent's own stream the other one has already been told. A handoff sends only the missing part, in four bounded sections: Conversation, Decisions, Work, Next. The receiving agent spends one short sentence acknowledging it. That is the entire overhead.
If you have ever worked on reconciliation systems, this is the same shape: you do not resend the ledger, you send the diff since the last watermark.
What it does not do
- It does not replace the agents.
- It does not proxy their APIs.
- It needs no API keys.
It drives the subscription CLIs you already have. One command, /bridge codex, and the agent you leave closes while the one you arrive in already knows your work.
The honest asymmetry
Not every agent accepts context the same way, and I think glossing over that would be dishonest.
- Claude and Codex take the delta through their own session hooks, so it lands inside the conversation.
- Grok and Antigravity get it in the opening prompt of the resumed session, because their runtimes offer no way to inject.
The knowledge is equal on every side. The session shape is not, and that difference is stated plainly rather than hidden.
The bug that reported success and carried nothing
Here is the failure that scared me most, and it was not in my code.
These agents store their sessions in undocumented internal files. Nobody owes you stability there. A point release renames one field and nothing looks broken: the binary still runs, auth still works, and the handoff reports success. Meanwhile every delta is quietly empty.
So bridge doctor got a check nobody asked for. It does not just check that an agent is installed and logged in. It opens each vendor's session files and confirms that the current version of the bridge can still parse them. That failure is invisible otherwise, until you notice that a week of handoffs carried nothing.
The lesson I keep from it: when you build on another product's internal files, "installed and logged in" tells you nothing. Test the data you depend on, not the binary that writes it.
Try it
It is a developer preview, tested and used daily, and open source. The write-up with install steps and the full list of honest limitations is on the project page: context-bridge on dogrubakar.com, and the code is on GitHub.
If you also juggle multiple agents, I would love to hear how you handle the context problem today.
Top comments (0)