Amp in NeuroSquadAmp,with a canvas around it
Run Amp in as many cards as you need, each with a real terminal. Amp has no hooks, so NeuroSquad brings a small plugin: it reports every turn, asks before third-party tools, and keeps each card on its own thread.
Official siteampcode.com- The real
ampCLI, your own Amp account - Nothing written to your Amp config
- Free, for Windows 10 and 11
It knows when Amp is done — and when it’s asking
Amp runs its agent on Amp’s server and has no lifecycle hooks. Each card loads a NeuroSquad plugin instead, and the plugin reports Amp’s own events as they happen.
agent.startplugin eventAmp starts on your prompt (agent.start). The card’s icon spins, the Squad status card counts it as working, and a waiting toast is taken down.
tool.call → ui.confirmplugin eventA tool from an MCP server installed in NeuroSquad is about to run. The plugin opens Amp’s own dialog, the card pulses and the toast names the tool. It waits for your answer.
agent.endplugin eventAmp finished handling the prompt (agent.end). The plugin passes its final answer on, so the journal, hand-off and agent_wait get it — the transcript itself lives on Amp’s server.
Click, and you’re there
A toast click switches to the workspace, flies the camera to that card and flashes it — even among dozens of agents.
Amp’s own permissions too
If your Amp settings ask before some commands, the thread’s awaiting-approval state marks the card as waiting, with the tool it waits on.
Helpers don’t count
Subagent threads (they have a parent thread) never move the card’s status — only the thread you talk to does.
Three things you’ll do with Amp
Replicas of the app, played step by step. When Amp needs your answer, the answer is yours.
Amp fixes a failing test, then wants a Linear tool from an MCP server installed in NeuroSquad — the plugin asks you first. The journal writes itself.
Everything Amp can do in NeuroSquad
Most of what Claude Code gets, rebuilt on Amp’s plugin API. Where Amp’s design rules something out, the card says so plainly.
- 01
Knows its state
A NeuroSquad plugin inside Amp reports every turn, so nothing is guessed from a quiet terminal.
A plugin instead of hooks
Amp loads plugins only from your project or your config folder, so each card gets its own config folder with the NeuroSquad plugin and copies of your own plugins.
Docs: A plugin instead of hooksagent.startmeans working,agent.end— finished.A notification that says what it needs
The toast leads with what Amp waits on — the tool and its input. One per agent; a waiting one stays until you answer; a click flies to the card.
Docs: A notification that says what it needsEvery card keeps its thread
Amp names its threads on its server (
Docs: Every card keeps its threadT-…). The plugin reports the id on every turn, and the next launch runsamp threads continue <id>. A deleted thread starts a fresh one by itself.The Squad status card
Every AI agent of the workspace on one card, with its status and turn time. Amp’s model comes from
Docs: The Squad status cardamp threads contextafter each turn; before the first one, from the mode it starts in.Custom cards can follow it
Cards built with the Card SDK see the same statuses and turn events, and read Amp’s final answers as the plugin reported them.
Docs: Custom cards can follow it
- 02
Works through the canvas
An arrow from the card is a set of tools. Draw it, and Amp can use what is at the other end.
Arrows are tools
An arrow to a browser, terminal, note or another agent gives Amp
Docs: Arrows are toolsmcp__neurosquad__browser_*,terminal_*,note_*oragent_*— Amp re-lists the tools as soon as the arrow is drawn.It can add cards itself
Docs: It can add cards itselfcanvas_spawn_cardopens a terminal, a browser, a note or a helper agent — another Amp included — next to it, already joined by an arrow.- Built by NeuroSquad
Canvas mode that holds
The plugin refuses Amp’s own
Docs: Canvas mode that holdsBash,shell_command,web_searchandread_web_pageand points it at terminal and browser cards. Plugin refusals win over--dangerously-allow-all, so it holds in dangerous mode too. - Built by NeuroSquad
MCP servers from the registry
Amp runs tools without asking by default. For servers you installed in NeuroSquad the plugin asks in Amp’s own dialog before every call — “Allow once” or cancel.
Docs: MCP servers from the registry Skills from skills.sh
An arrow to a skill card gives Amp
Docs: Skills from skills.shskill_loadwith each skill and when to use it.
- 03
Built for long runs
Threads that span hours, with context and credits in view.
Context meter, no Compact button
After each turn the card reads
Docs: Context meter, no Compact buttonamp threads contextand fills its meter. Amp compacts a thread itself on its server near the end of the window, and has no command for it — so there is no Compact button.- Built by NeuroSquad
Out of credits, clearly shown
A turn that ends with Amp’s own “out of credits” or “rate limit” wording puts the limit chip on the card. Credits have no reset time, so there is no countdown.
Docs: Out of credits, clearly shown Prompt queue
Line up the next messages: each goes out when the plugin reports the turn finished. Daily prompts too.
Docs: Prompt queueHand off to a fresh agent
Start a new agent beside it — Amp or any other CLI — with a summary of the work typed in, ready for you to send.
Docs: Hand off to a fresh agent- Built by NeuroSquad
Journal
Each finished answer is added to a connected note. The plugin keeps each turn’s answer and touched files on your machine, because the thread itself lives on Amp’s server.
Docs: Journal
- 04
Works in a team
Several Amp cards — or a mix of CLIs — on one project without stepping on each other.
Isolated worktrees
Give an Amp card its own git worktree, so parallel agents never edit the same files. A setup command and files to copy prepare it before Amp starts.
Docs: Isolated worktreesAdd a reviewer
One click starts an agent on a different CLI in the same worktree, with an arrow from the author and a prompt to review without editing anything.
Docs: Add a reviewerDangerous mode, per card
Starts Amp with
Docs: Dangerous mode, per card--dangerously-allow-all. Amp asks little by default, so it matters for your ownamp.permissionsrules — and it also skips NeuroSquad’s question for third-party tools.Roles
Reviewer, tester, researcher, architect, writer or your own: the role is typed in as a first prompt you can see and edit.
Docs: Roles- Built by NeuroSquad
Instructions, delivered
Amp reads AGENTS.md from your project, and your global AGENTS.md is copied into the card’s config folder. NeuroSquad’s own session instructions ride along hidden with the first prompt of each thread.
Docs: Instructions, delivered
- 05
Cost, control and access
What NeuroSquad can and cannot see of an agent whose model runs on Amp’s server.
- Not possible
No per-card cost — honestly
Amp bills turns on its server and keeps no usage log on your machine, so Amp cards are not in the Usage section. Your Amp account shows what they spent.
Docs: No per-card cost — honestly Budget brake, without counting
Past a workspace’s limit, every agent is stopped — Amp with Escape twice, its thread kept. Amp’s own spend is not counted towards the limit.
Docs: Budget brake, without counting- Not possible
No OpenRouter for Amp
Amp’s models are served by Amp; which provider answers is a setting of your Amp account. A card can’t be pointed at OpenRouter, and NeuroSquad can’t hand Amp a key.
Docs: No OpenRouter for Amp Dictation into the prompt
Hold a hotkey and speak. Speech is recognised on your machine and typed into the active agent’s prompt.
Docs: Dictation into the promptFrom your phone
Scan a QR code and the whole canvas opens in the phone’s browser: read Amp’s terminal, answer its question, send the next prompt.
Docs: From your phone
Who’s waiting — and why there’s no bill here
The Squad status card works for Amp as for any agent. The usage table stays empty on purpose.
The Squad status card
Every agent of the workspace with its status and turn time. The one waiting on you comes first, with the tool it asks about. Tap a status to filter.
- search-fixAmpAmp mediumAmp asks to run a third-party tool: mcp__ns-connected__linear__create_issueNeeds you1:17
- api-paginationAmpAmp mediumPaginate the orders APIWorking11:07
- csv-exportAmpAmp mediumAdd CSV exportWorking4:08
- docs-passAmpAmp mediumUpdate the READMEFinished6 min. ago
Usage & cost
NeuroSquad reads cost from logs a harness writes on your disk. Amp writes none: its agent runs on Amp’s server and is billed there.
The real Amp, plus a few arguments
NeuroSquad starts the same `amp` you run yourself, in a real terminal in the card. This is everything it adds:
Your Amp config is not written
The card’s config folder lives in NeuroSquad’s data folder and holds copies, never links. Your settings file is used as is.
Your account, your plan
Amp runs with the login it already has. Your NeuroSquad account is separate and never sees it; Amp’s server is Amp’s.
Found where you installed it
The install script’s amp.exe first, then npm’s — whose amp.exe is started directly, without a cmd shim in between.
Amp in NeuroSquad, answered
Give Amp a canvas
Free for Windows 10 and 11. Your code and your Amp login stay yours.