Problem & Solution

Running command-line coding agents day to day surfaces the same three problems repeatedly. NEXUS's design centers on one answer to each — in its own words, drawn from what NEXUS shows you on first launch and inside the app itself.

An agent shouldn't reset every time its shell closes

Working in a plain terminal, "the agent" is really just whatever process happens to be running in that shell. Close the shell, and its context, its persona, and anything it had learned about your project go with it — the next session starts from zero.

NEXUS separates the two: an agent is a named identity with its own persona and on-disk memory, and a shell is just where it happens to be running right now. As NEXUS's own first-launch tour puts it:

"Every shell runs as a named Agent with its own persona and on-disk memory. Rename it, duplicate it, come back next week — it's still the same agent, and its identity is injected into the terminal."

See Agents for how identity, persona, and autonomy are configured.

A workspace with several agents in flight is hard to hold in your head

Once more than one agent, session, and file are active in the same project, keeping track of what's connected to what — which session touched which files, which agents have been talking to each other, which notes belong to which piece of work — outgrows what fits in a sidebar list.

NEXUS derives a live map of the workspace instead of asking you to reconstruct it from memory:

"Your agents, sessions, files and Bits form a Neural Map you can navigate. Capture a Bit and it becomes a node; double-click any neuron to jump to the real thing."

See Neural Map for how nodes, connections, and drilling in work.

Choosing an agent shouldn't mean choosing a cloud

Many agent tools ask you to route your project through their own hosted service to get persona, memory, or coordination features — which means picking one vendor's agent also means accepting its cloud, its retention policy, and its lock-in.

NEXUS hosts whichever CLI agent you already use, on your own machine:

"Run any CLI agent — Claude Code, Codex, Gemini, your own. Everything stays on your machine; nothing is shared without your say-so. No Electron, no vendor lock-in."

A workspace's insights view makes the same point about its own data: "All on this device. Nothing is shared." See What is NEXUS? for the local-first architecture this rests on.

Where to go next

• Use Cases — concrete scenarios built on these three ideas.
• Agents and Neural Map — the guides for each.
• Local-First — the architecture behind provider-agnostic, on-machine operation.