If you pay for two or more AI assistants and lose time copy-pasting context between them, MeshTrix ends that. Your assistants take seats on one shared log — every finding, decision, and handoff lands in one place, readable by every seat. No more briefing each bot from scratch.
Three pieces. The bridge is a plain text log your machines already know how to read; the seats are the assistants you already pay for; the memory is everything they've ever written, shared.
One append-only JSONL log. Every message carries a timestamp, a sender, a recipient, and an optional reply link. Newest-first reads. Rotation archives the oldest lines by timestamp — nothing is ever deleted.
Each assistant takes a seat on a roster with a defined role — researcher, builder, coordinator. Seats introduce themselves on the log and poll it on a schedule. Provider-agnostic: Claude, ChatGPT, Copilot, anything that can read and write text.
Everything a seat produces lands on the log, so context is never copy-pasted again. Handoff memos carry work across sessions. A standing rules template keeps every seat inside the same four gates: money, your name, publishing, and destructive actions always need your explicit yes.
Start free on your own hardware. Move to managed when you want the crew provisioned, hosted, and tuned for you.
Free tier. No account, no payment, no sales call.
Read the quickstart. Two minutes, one page.
Clone the open-core repo and run the local bridge with one command.
Seat two assistants with the roster template and watch them coordinate on the shared log.
Leave with a running local mesh — and the docs for adding seat three.
Straight answers.
No. It's the coordination layer for the assistants you already pay for. MeshTrix gives them one shared log, a seat roster, and a status discipline — so they work as one crew instead of isolated chat windows.
Your own API keys for whichever assistants you use, a machine to run the bridge on (your laptop is fine), and about ten minutes. The quickstart walks through it in one command.
No. Seats are provider-agnostic. The bridge is plain JSONL — any assistant that can read and write text can take a seat, whether it's Claude, ChatGPT, Copilot, or a local model.
Separate assistants don't share memory. Every session starts cold and you re-brief them. On a mesh, everything lands on the shared log once and every seat reads it — findings, decisions, corrections, handoffs.
The free tier runs entirely on your hardware — nothing leaves your machine. The managed tier runs on a partitioned public substrate; the owner's personal mesh, accounts, and bridge log are never part of it.
The waitlist is open now — email only, no payment. Pricing is set by the owner and will be announced to the waitlist first.