memeri

The Terminal — run agents in the platform

Status: ✅ Current · Last reviewed: 2026-06-18

The Terminal page runs AI agents — Claude Code, Codex — right inside the platform, backed by a small local console daemon on your machine. No separate window; the agent runs in the browser.

Agents run on this computer. Your project data — jobs, memory, decisions, roadmap — syncs everywhere you sign in, but a running terminal is a live process on the machine that started it. Open Memeri on a different computer and you'll see all your project data, but that machine's Terminal will be empty (or "Console offline") until you start agents there. "Resume" below means a conversation picking up its context — not a process hopping between machines.

A chat is an agent

The core idea: each chat is a conversation, and a conversation IS an agent. There's no separate "agent" object to manage. You start a chat, name it, and that conversation carries its own identity, memory, and roadmap. Close it and resume it later — it picks up where it left off.

Solo chats and workspaces

You can run work two ways:

  • Solo chat — a single standalone conversation. The big green New button starts one.
  • Workspace — several chats grouped together as tabs, so related agents live side by side.

How the buttons map:

  • New (green) — start a standalone solo chat
  • + on a workspace tab — add a chat to that workspace
  • + at the end of the tab strip — make a new workspace

The pane header

Each chat pane has a header showing:

  • Its name (auto-named from the first response, editable)
  • A status dot — live / idle / stopped
  • A per-chat colour theme so panes are easy to tell apart
  • A copy control
  • A drawer toggle

The drawer

Open the drawer on any pane (the panel toggle in the header) for three tabs:

  • Roadmap — the agent's live plan: the Now / Next / Later steps it's working through
  • Context — what the agent is grounded on, plus anything you've attached
  • Memory — the project memory scoped to that chat (decisions, notes, recalled items)

This keeps each conversation's working state close at hand without cluttering the terminal itself.

Pop a chat out to its own window

Click the ⤢ pop-out button on a pane to detach it into its own OS window — perfect for putting an agent on a second monitor while you work elsewhere. The popped-out window carries the full experience: the live terminal, the colour-theme picker, and the same roadmap / context / memory drawer (opening the drawer simply grows the window, so nothing gets covered). The grid leaves a placeholder where it was; click ⤢ Pop back in (or just close the window) to return it. The agent keeps running the whole time.

Your Home is scoped to the active project

The Terminal Home follows the project you've selected (the project switcher, top-left):

  • Jump back in (chat history) and Saved workspaces show only the active project's chats.
  • Running agents stay visible no matter what — a live agent is never hidden. Agents bound to other projects drop into a dimmed "Other projects" group with a small project chip, so they stay reachable without cluttering your view.
  • A "Show all projects" toggle reveals everything at once.

So switching projects gives you a clean, focused terminal, and a brand-new project starts with an empty, uncluttered Home.

Native terminal ⇄ platform pane

You're not locked into the browser. You can switch freely between your own native terminal and the platform pane for the same live agent — run an agent locally (memeri attach), then continue it in the browser (or vice versa). Conversations persist and resume across both.

What it needs

The Terminal relies on the local Console daemon running on your machine (and the Tunnel, if cloud AIs need file access). Watch the Console / Tunnel status dots in the header — green means you're good. If the console is offline, the page tells you the one command to start it.

See Choosing your setup tier and Connect Your AI for getting those running.