Conversations are agents
Status: ✅ Current · Last reviewed: 2026-06-18
In Memeri, there is no separate agent-identity layer — a conversation IS an agent. This is the core primitive. There's no extra "agent" object to create, register, or keep in sync; you work with conversations, and each one is the agent.
The durable unit
Each chat / conversation is the durable unit of work:
- You name it — auto-named from the first response, editable any time.
- You resume it — close it and pick it back up later; it carries on where it left off.
- It carries its own memory and roadmap — the context, decisions, and plan for that line of work live with the conversation.
Because the conversation is the identity, you can run it in your own native terminal and continue the same live agent in the platform Terminal pane (or the other way around). See Using Memeri → The Terminal for how that works in the UI.
Naming
Names are stable and human-friendly. A conversation keeps its name across resumes and reconnects — so "the auth refactor chat" is the same thread today as it was yesterday. Rename it whenever the work shifts; renaming updates the live conversation's identity going forward.
Renaming a conversation doesn't rewrite history: events and updates already emitted keep the name they were emitted with. The rename applies to the live identity from that point on.
Why this matters
When every chat is a durable, named agent that carries its own memory, you stop losing work between sessions and between tools. You don't re-explain where you left off — you resume the conversation, and it already knows. And because conversations contribute to the same project record, what one learns becomes available to the next.
Cross-references
- Using Memeri → The Terminal — running conversations as panes, solo or in workspaces
- Working With Agents → Posting good job updates — how a conversation's progress becomes legible