Kimi Code in NeuroSquadKimi Code,with a canvas around it
Run as many Kimi Code agents as you need, each in its own card with Kimi’s real TUI. Kimi’s own hooks tell NeuroSquad when each is working, waiting on you or finished; arrows hand it terminals, notes and other agents; its sessions and tokens stay in view.
Official sitekimi.com/code/en- The real
kimiCLI, your own login - Nothing written to
~/.kimi-code - Free, for Windows 10 and 11
It knows when Kimi is done — and when it’s asking
A quiet terminal can mean “finished” or “waiting for your permission”. NeuroSquad doesn’t guess: each card’s config carries Kimi Code’s own hooks, and they report every turn.
UserPromptSubmithookYou press Enter. The card’s icon spins, the Squad status card counts it as working, and the toast from its last question is taken down.
PermissionRequesthookKimi asks before a command. The toast leads with Kimi’s own words — Running: rm -rf dist — and stays until you answer.
StophookThe turn is over. The next queued prompt goes out, the journal gets the answer, and a toast says it’s done. Esc sends Interrupt instead — also finished.
Questions and plans count too
When Kimi asks a question (AskUserQuestion) or wants its plan approved (ExitPlanMode), the card waits on you the same way.
Order is kept
Kimi fires some hooks without waiting, and a fast turn’s TurnStarted can arrive after its Stop. The card opens a turn on UserPromptSubmit, so it never sticks on “working”.
Your hooks stay yours
The card’s config is your own config.toml, verbatim, with NeuroSquad’s hooks appended. Hooks you set up still run.
Three things you’ll do with Kimi Code
Replicas of the app, played step by step. When Kimi asks, the answer is yours.
Kimi Code fixes a failing test through a terminal card, stops to ask before deleting a folder, and finishes — the journal writes itself.
Everything Kimi Code can do in NeuroSquad
Measured on Kimi Code 2.1.1. Most of it is Kimi’s own — hooks, MCP, sessions. Where NeuroSquad had to build something, or Kimi can’t do it, the card says so.
- 01
Knows its state
Kimi Code’s own hooks report every turn, question and failure, so nothing is guessed.
Finished and waiting are different
Each card’s config carries Kimi’s hooks:
Docs: Finished and waiting are differentUserPromptSubmitmarks it working,PermissionRequest— waiting on you,Stop— finished.A notification that says what it needs
The toast leads with Kimi’s words —
Docs: A notification that says what it needsKimi Code needs your permission: Running: …,Kimi Code is asking: …, or its plan to approve. It stays until you answer.Failures and interrupts
Docs: Failures and interruptsStopFailureends a turn that failed andInterruptone you stopped with Esc — the card never hangs on “working”.- Built by NeuroSquad
Sessions that survive a restart
Kimi creates its session id on the first message and has no flag to choose one. Every hook carries it, so the next launch resumes with
Docs: Sessions that survive a restart-S <id>— after/newtoo. The Squad status card
Every AI agent of the workspace on one card, with its model — from Kimi’s own request log, or your
Docs: The Squad status carddefault_modelbefore the first turn.
- 02
Works through the canvas
An arrow from the card is a set of tools, over Kimi’s own MCP client.
Arrows are tools
Each card’s
Docs: Arrows are toolsmcp.jsonadds NeuroSquad’s server, with the token in the environment. An arrow to a terminal, note, browser or agent gives Kimi those tools.- Built by NeuroSquad
The arrow is the consent
Kimi 2.1.1 reads permission rules but never applies them, so NeuroSquad answers Kimi’s own dialog for its tools with “Approve for this session”. Third-party tools still ask.
Docs: The arrow is the consent - Not possible
No live tool list
Kimi’s MCP client ignores
Docs: No live tool listtools/list_changed. So a card gets every NeuroSquad capability from the start, and the arrows decide at call time — a card spawned mid-session still works. Canvas mode that holds
Docs: Canvas mode that holdsBash,WebSearchandFetchURLare disabled in the card’s config, and aPreToolUsehook refuses them too — so--autocan’t lift it.MCP servers from the registry
Install a server from the official MCP Registry once, then give it to a card with an arrow. Kimi asks before every call to it; its secrets never leave NeuroSquad.
Docs: MCP servers from the registry
- 03
Built for long runs
Full contexts and quota limits are handled on the card.
Context meter
Tokens from Kimi’s request log against the model’s
Docs: Context metermax_context_size. The ring turns amber from 80%.Continues after a quota limit
A
Docs: Continues after a quota limitStopFailurewith Kimi’s quota error puts a countdown on the card, and it sends “continue” at the reset — “resets in 2h 15m” is read from the message.Prompt queue
Line up the next messages: each goes out on
Docs: Prompt queueStop, never while Kimi is still working. Daily prompts too.Journal and hand-off
The last answer is read from the session’s wire log: into a connected note, into a fresh agent’s first prompt, back to a lead waiting in
Docs: Journal and hand-offagent_wait.
- 04
Works in a team
Every card has a Kimi home of its own, so several agents never share a config or a session list.
A Kimi home per card
Docs: A Kimi home per cardKIMI_CODE_HOMEpoints to the card’s own folder: your config.toml verbatim plus NeuroSquad’s section, your MCP servers plus ours, links to your login. Your~/.kimi-codeis only read.- Built by NeuroSquad
No stop at “Trust this folder?”
Kimi asks before it starts in any folder it hasn’t trusted. The card writes the same trust record Kimi writes itself, because you already chose this folder.
Docs: No stop at “Trust this folder?” Isolated worktrees
Give an agent its own git worktree, so parallel agents never edit the same files. The worktree is trusted for Kimi the same way.
Docs: Isolated worktreesAdd a reviewer
One click starts an agent on a different CLI in the same worktree, with an arrow from Kimi and a prompt to review without editing anything.
Docs: Add a reviewerDangerous mode, per card
Docs: Dangerous mode, per card--auto— Kimi’s Never Ask mode. Turn it off and the next resume returns to your usual mode, even though a Kimi session remembers the mode it ran in.
- 05
Cost, control and access
What it spends, what it may spend, which model it runs on, and how you reach it.
Exact token counts
Every request is read from Kimi’s own wire log, subagents and compaction included, and credited to the card whose home holds the session. A model with no public price shows “no price”, never $0.
Docs: Exact token countsBudget brake
Past the workspace limit the card gets Escape — on an idle Kimi, Ctrl+C arms “press again to exit”. The session is kept; automatic prompts stop until you raise the limit.
Docs: Budget brakeOpenRouter models
A provider of the card’s own in its config reads the key from one environment variable — never in the arguments or on disk — and sends NeuroSquad’s attribution headers.
Docs: OpenRouter modelsOnly the maintained Kimi
The old Python kimi-cli is archived, and bare
Docs: Only the maintained Kimikimithere downloads an installer without asking. The card checks the version first and refuses to start that one.From your phone
Scan a QR code and the whole canvas opens in the phone’s browser: read Kimi’s terminal, answer its question, send the next prompt.
Docs: From your phone
Who’s waiting, and what it used
Two cards that answer the questions you ask most when several Kimi Code agents 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 what it asks. Tap a status to filter.
- coupon-fixKimi Codekimi-for-codingKimi Code needs your permission: Running: rm -rf distNeeds you2:34
- api-migrationKimi Codekimi-for-codingMove routes to the new clientWorking16:32
- deps-auditKimi Codemoonshotai/kimi-k2Check outdated packagesWorking5:25
- docs-passKimi Codekimi-for-codingUpdate the READMEFinished8 min. ago
Usage & cost
Read from each session’s wire log in the card’s own Kimi home, so every request belongs to exactly one card.
| Agent | Tokens | Cost | ||
|---|---|---|---|---|
| coupon-fix | 1,392,604 | no price | ||
| api-migration | 905,117 | $0.63 | ||
| docs-pass | 228,470 | $0.16 | ||
| Total | 112 | 2,526,191 | $0.79 |
The real Kimi Code, in a home of its own
Kimi Code reads everything from one folder and has no settings overlay, so NeuroSquad gives each card its own. This is everything it adds:
Nothing written to ~/.kimi-code
Your config is copied, not edited; your login, skills and plugins are linked. Deleting a card removes its generated files, never its sessions.
Your login, your plan
Kimi Code runs with the account it already uses — kimi login in a terminal is shared with every card.
Found where you installed it
npm’s kimi.cmd first, then the install script’s kimi.exe, from the current PATH. A Python kimi is refused, never run.
Kimi Code in NeuroSquad, answered
Give Kimi Code a canvas
Free for Windows 10 and 11. Your code, your login and your costs stay on your machine.