Written alongside the releases: why NeuroSquad works the way it does, what we measured, and what broke on the way.
Three things in this release: nsq keeps itself up to date and switches over only when no agent loses anything, `--online` takes the phone page out of your Wi-Fi, and `nsq provider add` puts agents on your own model servers.
Run Claude Code, Codex, OpenCode or any command side by side in one terminal dashboard, close it, and get called — with the question — when an agent needs you. Free, MIT, no account.
NeuroSquad has its own page on OpenRouter now. Not because we asked — because every request our agents send there says where it came from. Here is what that gives you.
Gemini CLI has a successor, and OpenCode has a second major version. Both arrived as closed or rewritten programs with their own rules, and both needed their own way in.
Three new agent CLIs, three very different programs: a fork of a CLI we already support, a client/server platform, and a desktop agent’s runtime in a terminal. Each needed its own way in.
Remote access used to give a phone the desktop app squeezed onto a small screen. In NeuroSquad 0.1.273 a phone gets a grid of your agents built for a thumb, and a computer’s browser gets Canvas and Grid like the desktop.
Until now, giving an agent a tool meant you drew the arrow. In NeuroSquad 0.1.270 the agent can look up what is available and connect it to itself, and nothing new is installed without your yes.
A helper stuck on a hard bug no longer has to wait for you to notice. In NeuroSquad 0.1.266 the lead agent can move it to a bigger model, another provider or another agent CLI, and the conversation goes with it.
Give three agent CLIs the same task, then go back and watch exactly what each of them did, with the numbers next to it. NeuroSquad 0.1.264 records a whole workspace, replays it on a timeline, and measures every run with two official cards.
An agent card used to be tied to the CLI it was created with. NeuroSquad 0.1.253 lets you switch it to another one in place, conversation included — plus templates you can drop into an existing workspace, and Graphify, a live map of your code.
A good agent setup takes an afternoon to get right: which agents, which cards, which arrows, which plugins and servers. NeuroSquad 0.1.247 lets you save it as a template, publish it, and lets anyone else rebuild it on their machine after one confirmation.
Agents spend turns grepping through code they could simply ask about, and write against library APIs as the model remembers them. NeuroSquad 0.1.230 adds two plugins for that: Code Graph, a map of your code, and Context7, documentation from today.
An agent forgets everything when its session ends. NeuroSquad 0.1.222 adds Memory: draw an arrow and the agent can save what it learns and find it again next week — kept on your computer, searchable by meaning. And Claude Code cards no longer get stuck on Working.
The biggest release so far: a native macOS app, a Grid mode for when the squad is just working, workspaces that live in WSL or on a remote machine, and a “needs attention” signal that goes away the moment you answer.
Some add-ons are neither tools nor instructions — they change what an agent’s shell prints or how it answers. NeuroSquad 0.1.160 calls them plugins: draw an arrow and they apply to the next command, remove it and they stop.
How much of your 5-hour and weekly windows is left, on every card — without NeuroSquad ever touching your login. And when one account runs dry, another is a click away, never an automatic switch.
Launching a CLI in a terminal is the easy tenth of an integration. The rest is statuses, sessions, tools and interruption — and every CLI needs a different way in.
Two releases in one morning opened the canvas to code we did not write. The hard part was not the features — it was keeping the promise that an agent reaches only what an arrow gives it.
An idle canvas with three agent cards was using a full CPU core. The culprit was a cartoon. Here is what we measured, what we changed, and the card that now sits on every canvas.
The installer went from 550 MB to 96 MB in a day, then got rewritten from scratch. What went wrong with the old one, and how updates now install without ever breaking the copy you have.
Three weeks after the first commit, the first public build. What problem it solves, and the three decisions everything else in the app is built on.