aider in NeuroSquadaider,with a canvas around it
aider has no hooks, no plugins and no MCP. NeuroSquad adds them from the inside: a small Python extension that loads with aider in each card, reports its turns and questions, and lets its model use the canvas — terminals, browsers, notes, other agents.
Official siteaider.chat- The real
aider, your own models and keys - Nothing written to
~/.aider.conf.yml - Free, for Windows 10 and 11
It knows when aider is done — and when it’s asking
aider’s only built-in signal, `--notifications-command`, fires the same way for “finished” and “asking you”. So NeuroSquad’s extension listens at aider’s own control points instead: sending to the model, asking a question, waiting at the prompt.
Coder.send_messageaider methodaider starts sending your message to the model. The card’s icon spins and the Squad status card counts it as working. Tool rounds and aider’s own lint and test fixes stay one turn.
io.confirm_askaider methodaider asks a question — run a command, add a URL, create a git repo. The card pulses and the toast leads with aider’s question and its subject, like “Run shell command? — rm -rf dist”.
io.get_inputaider methodaider is back at its > prompt. That, not a quiet terminal, ends the turn: the next queued prompt goes out, the journal gets the answer, a toast says it’s done.
Only real questions
Nothing is reported when aider answers itself (--yes-always, “don’t ask again”). The “what’s new” notice after an update is turned off with --no-show-release-notes.
One status everywhere
The card, the sidebar, the Squad status card, custom cards and the phone all read the same status.
Leaves no trace in your shell
The extension removes its own variables and restores your PYTHONPATH, so /run, /test and lint commands start clean.
Three things you’ll do every day
Replicas of the app, played step by step. When aider asks, the answer is yours.
aider fixes the parser and commits it, then suggests a shell command and asks before running it — the card waits for you, the journal writes itself.
Everything aider can do in NeuroSquad
aider brings no integration points of its own, so nearly all of this is built by NeuroSquad’s extension on aider’s own mechanisms. Each card says how — and the last group says where it stops.
- 01
Knows its state
aider’s own control points, patched from inside the process, report every turn and every question.
Finished and waiting are different
Docs: Finished and waiting are differentCoder.send_messagemarks the card working,confirm_askandprompt_ask— waiting on you,get_input— finished: aider is back at its prompt. Emulated from aider’s own control flow, not guessed from silence.A notification with aider’s question
The toast leads with aider’s words and their subject —
Docs: A notification with aider’s questionAdd URL to the chat? — https://…,Run shell command? — npm test. One per agent; a click flies to the card.Every card is its own session
aider keeps one history per folder, so two cards in one project would restore each other. Each card gets its own history files, and
Docs: Every card is its own session--restore-chat-historyafter a restart./clearand/resetstart a new session.The Squad status card
Every AI agent of the workspace on one card, with aider’s model from its last request (before that
Docs: The Squad status cardAIDER_MODELor your.aider.conf.yml) and the live time of its turn.Custom cards can follow it
Cards built with the Card SDK see the same statuses and turn events, and read aider’s last answer from the card’s session log.
Docs: Custom cards can follow it
- 02
Works through the canvas
An arrow from the card is a set of tools. The extension is the MCP client aider doesn’t have.
Arrows are tools — without a restart
An arrow to a terminal, browser, note or agent gives aider’s model
Docs: Arrows are tools — without a restartterminal_*,browser_*,note_*oragent_*. The extension re-reads the tool list before every request, so a new arrow works on the very next one.It can add cards itself
Docs: It can add cards itselfcanvas_spawn_cardopens a terminal, browser, note or helper agent next to it, joined by an arrow — an aider helper gets its first prompt once it reports it is at its prompt.Canvas mode
Docs: Canvas mode--no-suggest-shell-commandsand--no-detect-urlsswitch off aider’s only two ways to run a command and reach the web. Neither is an approval, so--yes-alwayscan’t bring them back.MCP servers from the registry
Servers you installed in NeuroSquad reach aider through a second client. Each call asks first through aider’s own
Docs: MCP servers from the registryconfirm_ask— “Run third-party tool?” — with an explicit yes.Skills from skills.sh
An arrow to a skill card gives aider’s model
Docs: Skills from skills.shskill_loadthrough the same bridge, live on the next request.
- 03
Built for long runs
Full contexts, rate limits and long queues are handled on the card.
Context meter and Compact
The ring shows the last request’s prompt against the model’s window from aider’s own model info. aider has no manual compaction — the extension adds
Docs: Context meter and Compact/compact, running aider’s own summariser over the whole chat.Continues after a rate limit
After aider’s own retries give up, the final error reaches the card. A reset time like “try again in 2h 15m” is read from it, and “continue” goes out then.
Docs: Continues after a rate limitPrompt queue
Line up the next messages: each goes out when aider is back at its prompt — exactly, not after a pause. Daily prompts too.
Docs: Prompt queueHand off to a fresh agent
Start a new agent beside it — aider or any other CLI — with a summary of the work typed in, ready for you to send.
Docs: Hand off to a fresh agentJournal
Each finished answer is added to a connected note. The text comes from the card’s session log, with the tool calls cut out.
Docs: Journal
- 04
Works in a team
aider next to other aider cards, or next to Claude Code and Codex, on one project.
Isolated worktrees
Give an aider card its own git worktree, so its auto-commits never collide with another agent’s edits. A setup command and files to copy prepare it first.
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 reviewerLead or helper
An arrow to another agent gives aider
Docs: Lead or helperagent_sendandagent_wait— or makes it the helper. Tested live: an aider lead spawned an aider helper and got its answer back.Dangerous mode, per card
Docs: Dangerous mode, per card--yes-alwaysanswers aider’s questions for you. aider still insists on an explicit yes before a shell command — its own rule, kept.Roles
Reviewer, tester, researcher or your own: the role is typed in as a first prompt you can see and edit, and never re-sent behind your back.
Docs: Roles
- 05
Cost, control and access
What it spends, what it may spend, which model it runs on, and how you reach it.
Usage and cost per request
aider prints tokens and cost and forgets them. The extension logs each request with the provider’s own usage block — cache included — and counts the side calls aider leaves out of its totals, like commit messages.
Docs: Usage and cost per requestBudget brake
A limit per workspace. Past it, aider gets one Ctrl+C — which only interrupts, it doesn’t quit — and automatic prompts stop. Raise the limit and it’s lifted.
Docs: Budget brakeOpenRouter models
Run a card on an OpenRouter model with NeuroSquad’s key. It goes only into that card’s environment, and is re-applied after aider loads your
Docs: OpenRouter models.envfiles, so the card bills the key you chose.Dictation into the prompt
Hold a hotkey and speak. Speech is recognised on your machine and typed into aider’s prompt.
Docs: Dictation into the promptFrom your phone
Scan a QR code and the whole canvas opens in the phone’s browser: read aider’s terminal, answer its question, send the next prompt.
Docs: From your phone
- 06
How it’s done — and where it stops
What NeuroSquad has to build around aider, and what it deliberately doesn’t.
A Python extension, loaded by Python itself
Emulated. The card puts NeuroSquad’s folder first on
Docs: A Python extension, loaded by Python itselfPYTHONPATH; itssitecustomize.pypatches aider as it loads. Every patch is guarded — if a new aider renames a method, only that feature switches off.Tools through a text protocol
aider sends its model no tools at all. So the model writes
Docs: Tools through a text protocol<neurosquad_call tool="…">in its reply, and the extension runs it and sends the result back. Strong models follow it; a weak one may describe a call instead of making it.Ctrl+C lands when bytes arrive
On Windows, aider notices an interrupt only when the next bytes of the reply come in. A provider that stalls without streaming holds the interrupt until it answers.
Docs: Ctrl+C lands when bytes arriveNo AGENTS.md — on purpose
Impossible natively: aider reads no instructions file by itself (conventions only with
Docs: No AGENTS.md — on purpose--read). NeuroSquad doesn’t slip AGENTS.md in, because that would change what aider sends for everyone.Installed by you, not the wizard
aider is a Python app (pip, uv, pipx). The setup wizard installs only through npm and winget, so it leaves aider to you. The card finds
Docs: Installed by you, not the wizardaiderwherever it is on PATH.
Who’s waiting, and what it cost
Two cards that answer the questions you ask most when several aider cards run at once.
The Squad status card
Every agent of the workspace with its status, model and turn time. The one waiting on you comes first, with aider’s question. Tap a status to filter.
- coupon-fixaidergpt-5Run shell command? — rm -rf distNeeds you1:17
- api-migrationaidergpt-5Move routes to the new clientWorking11:07
- db-indexesaiderdeepseek-chatAdd the missing indexesWorking5:25
- docs-passaiderdeepseek-chatUpdate the READMEFinished3 min. ago
Usage & cost
Read from each card’s session log — one line per request, with the provider’s own token counts. The rows always add up to the total.
| Agent | Tokens | Cost | ||
|---|---|---|---|---|
| coupon-fix | 412,804 | $0.96 | ||
| api-migration | 305,117 | $0.71 | ||
| docs-pass | 88,430 | $0.03 | ||
| Total | 74 | 806,351 | $1.70 |
The real aider, plus a few options
NeuroSquad starts the same aider you run yourself, in a real terminal in the card. This is everything it adds:
Nothing written to your config
~/.aider.conf.yml, .env files and model settings are never touched. The options above only override the options they name.
Your models, your keys
aider runs on whatever model and key it already uses. Your NeuroSquad account is separate and never sees it.
Secrets stay out of files
The hook URL, the MCP token and any OpenRouter key travel only in that one process’s environment — never on disk, never in arguments.
aider in NeuroSquad, answered
Give aider a canvas
Free for Windows 10 and 11. Your code, your keys and your costs stay on your machine.