Context should follow the work.
People are starting to use different AI agents for different jobs. Those agents may work on the same problem, but they do not share a reliable understanding of it.
People are still the bridge.
A customer problem may begin in support, be discussed in a meeting, researched in another tool, and end up with a coding agent. Today, a person has to carry the useful details through every step.
They copy messages, rewrite summaries, repeat decisions, and explain what already failed. The agent receives the latest request, but not the story behind it.
Keeping transcripts is not enough.
An agent needs more than a pile of old conversations. It needs to know the current goal, what changed, which decisions still matter, what was tried, and what remains unfinished.
Termyte is being built around the work itself. Past conversations are useful evidence, but the current state of the work should guide what happens next.
Give each agent only what it needs.
The goal is not to send everything to every agent. Termyte should find the right piece of work, keep its history and current state connected, and prepare a short briefing for the agent taking the next step.
Coding-agent continuity
A local open-source alpha that tests this idea across Claude Code, Codex, and OpenCode.
Context across every agent
Work that can move through support, research, meetings, engineering, and operations without being explained again each time.
What we still need to prove.
- That this problem happens often enough to need a separate product.
- That structured work state helps more than transcripts or handoff files.
- That people trust the product to capture and share the right information.
- That a coding-agent product can grow into wider company workflows.