AgentSpace
liveRun a crew of AI agents in a pixel-art office — assign the work, watch it happen, get it delivered.
Hire AI workers.You just manage.
AgentSpace sets up an AI office on your machine: your crew takes the task, writes the code, runs the tests and reports back with proof. You set the goal — the crew handles the technical work.
.dmg · macOS 13+ · Apple Silicon · v0.2.43
.exe · Windows 10/11 · 64-bit · 130 MB · v0.2.43
AppImage · Linux · x86_64 · 156 MB · v0.2.43
Intel Mac? Download the x64 build (157 MB)
Different platform?macOS (Apple Silicon)Windows 10/11Linux (AppImage)
- macOS 13+ · Apple Silicon
- Windows 10/11 · 64-bit
- Linux · x86_64
LIVE DEMO — not a videothis site was built with itAgentSpace v0.2.43
Muratify AgentSpacev0.2.43
loading the live demo…
Muratify desktop
Three products. One office.
AgentSpace sets up the office; AgentShot adds your screen, AgentVoice adds your voice — and all three are managed from the same Muratify account.
AgentShot
liveA screenshot studio — capture, frame and turn it into something shareable in one move.
AgentVoice
liveSpeak Turkish and the text lands right where your cursor is — in any app you're in.
Mobile app · your own private network
AgentSpace in your pocket — not watching, running it.
The phone is not a second product, it's the office's second screen: the same crew, the same board, the same log. You hand out work by voice, open a task and delegate it to an agent, clear a pending approval with one tap. Move through the tabs in the frame on the right, then open Device to see exactly which door the phone comes through.
- Give work by voice: the audio is transcribed on the Mac — your engine key is never copied to the phone
- Open a task from the phone and delegate it to an idle agent — the work starts in a real terminal
- Send a photo or a screenshot from the camera and attach it to your command
- Approve or reject what's pending, stop a run that's going wrong — the crew never waits on you
- It answers with the desktop window closed: the office is assembled in the Mac's main process
- In a café, on the road, in another city — the phone still comes in over your own private network
The interface in the frame is the product's own mobile app: the tabs, the screens and the Device card are all its own. What the phone cannot do stays visible too — no mouse/keyboard, no file deletion, no settings changes on mobile.
Network and data
Your data is on your disk.
Your code, your terminal output, your agent reports and your engine keys stay on your own machine. What follows isn't the slogan — it's the itemised version of it.
STAYS ON YOUR DISK
- Your code and your files
- Terminal output and session logs
- Agent reports
- Images you upload from the phone
- The memory files the agent writes itself
- Your engine keys
SYNCS TO YOUR ACCOUNT
- Your account and licence
- Your roster — agents and their roles
- Your board cards: title, description, status, assignee
- Memory rules you approve and promote by hand
- Product analytics with no open-text field (can be switched off)
The left column is files on your disk — there is no route in the product that ships them to us. The right column is the office record itself: short, countable, and written out here one by one. It sits in columns because it does not fit in a slogan.
What stays on your disk, and which door the phone comes through.
The honest version, with the mechanism named — not adjectives. Every sentence below is backed by a line of the product's code.
Your code and files stay on your disk
Agents run in real terminals on your own machine. Terminal output, session logs and agent reports are files on that disk, and the phone reads them straight from your Mac. What syncs to your AgentSpace account is the office record: board cards, your roster, and the memory rules you promote by hand.
The phone comes in over a private network
Being in a café, on the road or in another city changes nothing: the mobile gateway binds only to your Tailscale address (100.x), and that address is not routed on the public internet. Binding to every interface (0.0.0.0) is refused in code — a setting cannot punch through it. We measured it: from the same Wi-Fi, the same port answers with a refused connection.
Transport runs inside a WireGuard tunnel
Traffic between phone and Mac is encrypted by Tailscale's WireGuard tunnel. There is a per-device token at the door: a request without one gets a 401, you revoke a device with one click from the desktop, and the kill switch cuts every mobile route at once.
What we are not claiming
Not "nothing ever leaves your machine". You bring your own engine with your own key, so your model traffic goes straight to the provider you chose — not to us. And product analytics has no open-text field at all: a prompt, a file path or a key cannot pass the schema, and you can switch analytics off in Settings.
Bring your own engine
Your engine, your subscription, your bill.
AgentSpace doesn't resell tokens. You connect the AI CLI you already pay for, the key stays on your disk, and the plan price covers the office — not the model.
Running today · 8 engines
Assign one to any worker and it spawns in that worker's pane. This list isn't written on this page — it's generated from the app's own engine catalog.
- Claude Code
- Codex
- GitHub Copilot CLI
- Goose
- Gemini CLI
- Qwen Code
- OpenCode
- Crush
Model providers
Not engines of their own — providers you can point Codex at with your own key. Listed under the engine that actually runs them rather than padding the count.
- Groqvia Codex
- DeepSeekvia Codex
- Kimi K2/K3 (Moonshot — model)via Codex
On the roadmap · 13
Not wired up yet. They're dimmed on purpose: nothing on this wall lights up unless it really runs today.
- Kiroroadmap
- Grok CLIroadmap
- Cursor CLIroadmap
- Antigravityroadmap
- Kimi CLIroadmap
- Droidroadmap
- Amproadmap
- Clineroadmap
- Continueroadmap
- Charmroadmap
- Muse Coderoadmap
- oh-my-piroadmap
- Hermes Agentroadmap
- Any CLI agentthe list isn't closed
A new engine on this wall means new code in the product. The running list is generated from the app's engine catalog and a build gate turns red the moment the two drift apart — including if something here quietly starts working and is still shown as "roadmap".
Accounts & usage
Two subscriptions, one office — a limit doesn't stop the work.
Agents never hide which subscription they're burning: the active account is written in the title bar and the tokens spent sit above the pane. When the active account hits its limit, the work doesn't park for five hours — it continues on the idle account, in the same conversation.
agentspace — accounts & usageaccount-a@example.com
Settings → Accounts
- account-a@example.comactivesigned innew panes run on this account
- account-b@example.comsigned inSwitch to this account
Settings → Accounts → "switch to the idle account when the limit hits" — off by default; you turn it on.
Above the pane
This session
0.42M tokens
would cost ~$2.10 on the API
covered by your subscription
All sessions4.02M tokens
By model
- high-end model72%
- fast model28%
running · account A
Step 1 of 4
Running
The agent runs on the first account. The card above the pane shows the real tokens burned in this session and what they would have cost on the API — read from the engine's own ledger, not estimated by us.
The numbers come from the engine's own session files; the card never computes or invents a figure, and it prints "unpriced" instead of a fake $0.
What this section does not promise
No percentage quota bar
Engines don't publish how much of your quota is left, so we don't draw a bar for it. You get what's actually measurable: tokens spent, the API equivalent, and the hour work resumes.
Open panes keep their account
A switch applies to panes opened (or resumed) afterwards. A process that's already running keeps the account it started with — the app says this on the same screen.
First sign-in is the engine's own flow
Adding a second subscription runs the engine's own login once. After that, moving between signed-in accounts is a single click.
No account sharing
There is no button that exports or imports a token. The only supported case is your own second subscription.
Sign-in state is read from the engine's own status command, never from a local "we started a login" optimism — an account row can only claim to be signed in if the engine says so.
Category comparison
The strongest? Count it: 8 axes, 8 ✓.
The columns are categories, not brands — each is good at its own job; this table doesn't trash anyone, it counts scope. Category cells summarize today's typical behavior; individual products may differ. Every ✓ in the AgentSpace column has its counterpart on this page — the clips are recorded from the real product.
| AgentSpace | IDE agents | Worktree orchestrators | Chat assistants | |
|---|---|---|---|---|
| Live officeThe crew sits in a pixel-art office, at their desks; who's working, who's idle — at a glance. | — | — | — | |
| Persistent identity + memoryNamed workers with roles; they remember decisions and preferences and pick up where they left off. | Partial | — | Partial | |
| Orchestration + delegationDrop a task on the board — by voice if you like; the lead splits it, the crew runs, work closes with proof. | Partial | Partial | ||
| Parallel isolated workAgents run at the same time, each in its own named terminal session — a real shell, real CLI output. | Partial | — | ||
| Mobile accessThe office in your pocket: watch what your agents are writing, live from your phone. | — | — | ||
| Turkish supportThe interface and voice commands speak Turkish too. | Partial | — | ||
| Bring your own agent (BYOA)Connect the AI CLI you already pay for; the key stays on your disk. | Partial | — | ||
| Your data on your diskWorkspace, memory and keys live on your own machine. | Partial | — | ||
| Open source? | No — the code is closed; your data stays on your disk, and the building process is live on air, in the open. Watch it being built | |||
Category columns name no products and summarize today's typical behavior — these categories evolve fast, and individual products may deviate from the generalization. The AgentSpace column is no generalization: every ✓ is backed by real product recordings on this page.
How it works
From a prompt to a report.
Say the goal; the rest is the crew's job. Three steps and the work closes.
- 01
Say it
Write or speak the goal — “redesign the landing page”. You don't work out the technical detail.
- 02
The crew works
The task lands on the board, the lead splits it into subtasks and the agents run in their own terminals. You watch it live.
- 03
Close it with proof
The job closes with a result report and evidence — “typecheck + build green”, documented.
From the screen
Click in for a closer look.
Every card below is a recording of the real product — no mockups, no staging. Click a card: the clip opens; arrow through, press Esc to close.
Terminal
There's an identity at the end of the split.
Worktree orchestrators and terminal-only tools split the screen too. The AgentSpace difference: every split ends in a worker you call by name, with an office behind them.
Split without limits, drag, group, collapse.
The recording is not an edit — it's the product: split right, split down, drag the divider — every pane runs its own shell. Collapse the office group and the terminals take the full width.
- Four-way splitsEvery pane splits right, left, down and up; the tiling tree has no limit.
- Resize by draggingPull the divider; neighbouring panes re-flow in real time.
- Tab grouping + collapseDrop a tab onto another group's body to join it; collapse a group to give the screen to the work.
agentspace — splitting terminals
Office → terminal
Click in the office — the agent's named terminal appears.
Panes aren't called “shell 3” here. Click a worker's sprite in the office and a terminal carrying that agent's own name opens, with the command box now addressing them.
- Named terminalsThe pane title is the agent's name — you see who works where at a glance.
- A terminal per agentEvery worker runs in their own pane; sessions never bleed into each other.
- One-click bridgeClick the sprite; the agent's terminal and command line are waiting — you orchestrate by name, not by pane number.
agentspace — office to terminal
Plenty of tools split a terminal: worktree orchestrators, terminal-only multiplexers. Everyone splits. Nobody puts an agent with an identity at the end of the split — one whose name you know, who has a face in an office, whose terminal appears when you click them. In AgentSpace a split isn't a layout trick; it's your team's seating plan.
Integration Hub
You grant the connection. The agent does the rest.
Connect Sentry, GitHub, Supabase and more once from Settings; your agents then work on those platforms for you. No more digging through dashboards to find where a setting lives or how to update it.
- Connect once. Paste a key under Settings → Integrations and test it. New agent panes come up with that service's tools ready.
- The agent asks first. When a job needs a service, the agent queries the bridge: what's connected, with what scope, for which environment. It never guesses.
- Secrets never surface. Keys live in an encrypted vault; the agent's status query carries no secret — not even a masked one.
- Dev and prod stay apart. Connections are environment-scoped: production keys take an explicit flag, so a dev key never quietly does prod work.
Today's catalog
SupabaseGitHubSentryStripePostHogHostingerCoolifyVercel
The agent doesn't guess — it's straight with you in all three states
CONNECTED
Does the job itself
“Sentry is connected — I went through the latest errors myself: 3 new ones; I opened a board task for the most critical.”
CONNECTED · SCOPE SHORT
Names the missing permission
“GitHub is connected, but the key lacks ‘Issues: Read and write’. Add that to the token and I'll open the issue myself.”
NOT CONNECTED
No pretending — it shows the way
“Sentry isn't connected. Connect it under Settings → Integrations and I can list and summarize the latest errors for you.”
That “connect it and I'll do X” list isn't invented: the agent reads capability cards derived from each service's real tool list.
Who it's for
Who reaches for AgentSpace?
The common thread: anyone who wants to delegate the technical work and keep the direction.
Solo founder
Ship a product end to end without writing code: the crew does research, development, testing and the report; you set the direction.
Developer spreading the load
Split the boring, parallel work (refactors, test writing, migrations) across the crew; you stay on the architecture and the hard parts.
Small-team lead
Run an AI crew alongside your human team: hand out tasks on the board, follow progress in one window.
Researcher / analyst
Give competitor sweeps, data gathering and summarising to the agents; sources and tool calls come back marked.
Content & ops
Delegate repetitive production and editing flows; you handle quality control and sign-off.
Manager on the move
Leave the desk — follow what your agents are writing from your phone in read-along mode.
Proof
Not a concept — a working system.
Muratify's work, this page included, happens in this office every day.
muratify — the team office
Muratify's own crew — these agents built this page too.
Our own tool
AgentSpace is the system running at Muratify every day: the lead splits the work, the crew runs it, review happens live. Even this site was built with it.
First ones in
No other product gives you a genuinely working crew, gamified management and a non-coding founder at the same time.
Visual & gamified
Agents with characters, an office metaphor, live speech bubbles — managing it is enjoyable, not a dry dashboard.
“Even if you don't write software, build your own AI company visually and run it like a boss.”
Proof
We don't claim it. We show it.
Every line below ends in a source you can open — you don't have to take our word for any of it.
262+ hours on camera
52 streams over 112 daysThis product wasn't written behind closed doors. From day 1 the work has been live on stream, and the whole archive is still public — open any of it.
Open the stream archive20 updates in the last 30 days
~1 release a dayThree desktop apps, every release shipped with public notes. This number isn't typed by hand — it is read from the release feed, so it cannot go stale.
See the release historyWe build this app with this app
in 112 days45 AI agents · 2.960 tasks · 2.888 commits. These come from our own board — but you can watch that board and those agents on screen in the streams.
Watch what it looks likeReported bug → the release that closed it
no names, just links- 24.08the «⚠ N» badge read like an error counterclosed in 0.2.41
- 24.08the engine would not start with a long configurationclosed in 0.2.41
- 28.08the lead ran the work itself instead of dispatching itclosed in 0.2.42
Each line opens the «fixes from customer reports» item in that release's public notes. No reporter's name, handle or message appears anywhere — we don't quote anyone without permission.
None of these figures is typed by hand: they were measured at source on 2026-09-07 — the stream archive, our own task board and the repository history. The archive hours carry a «+» because a stream still on air is not counted yet. The tempo line is read live on every load.
Price
Basic, Pro, Ultra.
LAUNCH OFFER
50% off your first 3 months
Enter LAUNCH50 at checkout — the amount halves instantly.
LAUNCH50Use the discount
The discount applies across the Muratify AgentSpace ecosystem — AgentSpace plans plus AgentShot and AgentVoice. It does not apply to ChatFlow plans.
After the discount ends the plan renews at its normal price; you can cancel any time — cancelling also ends the discount. The 14-day refund guarantee also covers purchases made with the coupon.
Basic
Solo start — the first step
Discounted price: $7.50Billed monthly · cancel any time
LAUNCH5050% off first 3 months · then $15/mo
Start with Basic- 3 AI workers (agents)
- 1 workspace
- Bring your own engine — Claude Code, Codex, Groq, DeepSeek, Kimi
- 2 connected integrations
- Community access
- Community support (Discord)
Pro
RecommendedFor serious daily use
Discounted price: $14.50Billed monthly · cancel any time
LAUNCH5050% off first 3 months · then $29/mo
Start with Pro- Unlimited AI workers
- Unlimited workspaces
- Bring your own engine — Claude Code, Codex, Groq, DeepSeek, Kimi
- All integrations
AgentVoice included
AgentShot included- Early access
- Email support
Ultra
The whole ecosystem
Discounted price: $24.50Billed monthly · cancel any time
LAUNCH5050% off first 3 months · then $49/mo
Lifetime · one payment ↓Discounted price: $149.50- Everything in Pro, plus:
- Priority email + founder-direct onboarding
- VIP community space
- Beta channel
- Team seats (soon)
- Token-writer credit (soon)
Sign in with your account — the plan you picked and the same price are waiting at checkout; LAUNCH50 is applied automatically.
Prices in USD. You buy from your Muratify ID account portal — one login unlocks every plan and app. AgentVoice and AgentShot are included with Pro and Ultra.
Supported engines
Runs on your own engine subscription — token costs are not part of the plan price, and both the key and your privacy stay with you.
- Claude Code
- Codex
- GitHub Copilot CLI
- Goose
- Gemini CLI
- Qwen Code
- OpenCode
- Crush
- Groqvia Codex
- DeepSeekvia Codex
- Kimi K2/K3 (Moonshot — model)via Codex
…plus 13 more engines on the roadmap — listed separately on the product page, never shown as running.
LIFETIME · PAY ONCE
5 owners · next license #6Pay once — yours forever.
No subscription, no renewal, no surprises. One payment; the plan stays on your account.
Ultra · Lifetime
Discounted price: $149.50LAUNCH5050% off the one-time payment
Buy lifetime — License #614-day refund guarantee — coupon and lifetime purchases included
Against the Ultra yearly plan
yearly $399lifetime
- You pay today
- $199.50
- $149.50
- Every year after
- $399
- $0
- Total over 2 years
- $598.50
- $149.50
You pay $449 less over 2 years — lifetime pays for itself in 9 months.
Desktop app, agents, memory, skills and every update — forever. Cloud sync and mobile stay on subscription.
What's in each plan?
| BasicDiscounted price: $7.50 | RecommendedProDiscounted price: $14.50 | UltraDiscounted price: $24.50 | |
|---|---|---|---|
| Your AI crew | |||
| AI workers (agents) | 3 | Unlimited | Unlimited |
| Workspace | 1 | Unlimited | Unlimited |
| Bring your own engine — Claude Code, Codex, GitHub Copilot CLI, Goose, Gemini CLI, Qwen Code, OpenCode, Crush, Groq, DeepSeek, Kimi K2/K3 (Moonshot — model) | |||
| Integrations | |||
| Connected integrations | 2 | All | All |
| Included products | |||
| AgentVoice — voice dictation | — | ||
| AgentShot — screenshot studio | — | ||
| Early access | — | ||
| Support & community | |||
| Support | Community (Discord) | Priority email | |
| Community access | |||
| VIP community space | — | — | |
| Beta channel | — | — | |
| Coming soon | |||
| Team seatssoon | — | — | |
| Token-writer creditsoon | — | — | |
Value breakdown
If you bought everything in Pro separately
- Basic — Solo start — the first step3 AI workers (agents) · 1 workspace$15/mo
- AgentShot — Capture the screen, mark it up, share itAgentShot included$24 one payment
- AgentVoice — Unlimited local dictation on your own machineAgentVoice included$9/mo
- Piece-by-piece total$24/mo + $24 one-time
Pro $29/moonly $5 a month more than buying separately
That difference buys these lines — only in Pro:
- Unlimited AI workers
- Unlimited workspaces
- All integrations
- Early access
- Email support
- 14-day refund guaranteewe try to solve the problem together first; if we cannot, we refund within the first 14 days.
- Cancel any timeNo commitment, no hidden fees — you stop the subscription from the account portal in one click.
- Your data stays yoursBYO-key: your agents run with your key, on your machine. Token costs and privacy stay with you.
FAQ
Frequently asked
What problem does AgentSpace solve?
It lets a one-person team coordinate several AI agents — hand out work, watch it live, get it delivered with proof. The agents do the technical work; you manage it like a boss.
How is it different from IDE agents?
IDE agents give you one agent inside one editor session; AgentSpace runs a TEAM. Every agent has a persistent identity, its own role, its own real terminal and memory that survives between sessions; work lands on a task board that you hand out and watch live. The unit of work is “my crew ships this”, not “write me some code”.
Does it work together with Claude Code?
Yes — Claude Code is one of the engines AgentSpace runs today. You assign it to a crew member and it opens as a real session in that member's own pane, so you watch every command and every line of output. It runs on your own subscription or key; AgentSpace never swaps the model on you.
Which CLI agents and engines are supported?
The engines that actually run today: Claude Code, Codex, GitHub Copilot CLI, Goose, Gemini CLI, Qwen Code, OpenCode, and Crush. Underneath them you can also point Codex at provider models with your own key: Groq, DeepSeek, and Kimi K2/K3 (Moonshot — model). The list is not hand-maintained on this page — it is generated from the app's own engine catalogue, and the engine wall above always shows the current state.
Does it use my own API key?
Yes — BYO-key. Your agents use your own CLI key or subscription; token costs and privacy stay entirely with you. A step-by-step key wizard helps during setup.
How do I keep token costs under control?
Your token bill comes from your own AI provider, not from AgentSpace — we sell no tokens; the plan pays for the office, not the model. Limits, model choice and spend stay in the CLI subscription you already manage. You also decide which agent runs on which engine, so heavy work can go to a strong engine and routine work to a cheap one.
How many agents can run at the same time?
Your plan sets the crew size: Basic comes with a limited number of agents and a single workspace, while Pro and Ultra make both unlimited. How many run at once is up to you — each agent works in its own pane, so queueing them or starting them in parallel is your call. In practice the ceiling is your machine, not your licence.
Can it break my existing repo?
Agents work on your machine, in the workspace you point them at, in REAL terminals — every command and every line of output scrolls in front of you, with no hidden channel. You scope what they may do through the task itself, and starting agents in bulk needs your approval. As with any serious work, version control is the safety net: you review the diff when you take delivery.
Where does my data go?
Your code, your files and the memory files your agents write stay on your disk, and the agents work in real terminals on your own machine. Because you bring your own key, your model traffic goes straight to the provider you chose, not to us. What syncs to the cloud is the office record: your Muratify account and licence, your roster, your board cards (title, description, status, assignee) and the memory rules you promote by hand — never your code, your files or your terminal output.
Is the code open source?
No — the code is closed; your data is on your disk, and the build process is open on a live stream. We build the product with our own agent crew and share that process openly; closed source does not mean your data sits with us.
Do I need to know how to code?
No — that's exactly what AgentSpace is for. Talk or type, build your team visually, click to manage your agents. The technical work is theirs.
Does it run on Windows? Which platforms are supported?
macOS (Apple Silicon) and Windows 10/11 (64-bit, early access), plus a mobile read-along mode. Both installers are on the download page.
Can I use it from my phone?
Yes — a mobile companion mode mirrors your desktop office on your phone. You see who is working on what and where each task stands, read agent output in read-along mode, and answer pending permission prompts from the phone. The desktop app is what does the work; the phone is for following it and approving.
Does it support Turkish?
Yes — the interface is Turkish and the voice commands were built for Turkish specifically (including a Turkish morphology layer and end-of-sentence detection). You brief your agents in Turkish and get their reports in Turkish. We build the product itself while using it in Turkish.
How do I download AgentSpace and get started?
Get the app from the download page (muratify.com/download) — macOS or Windows — install it and sign in with Muratify ID; your first AI crew is up in a few minutes. Purchases and licences live in the same Muratify account; one login opens all three apps.
What's the difference between Basic, Pro and Ultra?
Basic is the solo starting point with a limited number of agents and workspaces. Pro gives you unlimited agents and workspaces and also unlocks AgentVoice and AgentShot (the hero bundle). Ultra adds founder-direct support, the VIP community and beta access on top of the same product.
Can I get a refund?
Yes — there is a 14-day refund guarantee with no exceptions: purchases made with a campaign code (LAUNCH50) and lifetime purchases are covered too. If the product does not do the job you expected, write to us within the first 14 days; the request goes through support first, we try to solve the problem together, and if we cannot we refund your payment. The guarantee applies once per account.
Which plan has the lifetime option?
Only Ultra. Pro has no lifetime option: while a one-time payment stayed cheaper than a year of the subscription, the yearly plan made no sense — so lifetime now lives in a single tier, at the top end. Basic and Pro are monthly or yearly.
How does it relate to the other products?
AgentVoice, AgentShot and ChatFlow — every product we ship — is built by the agent crew inside AgentSpace. This page included.
Get started
Stop writing.Start managing.
Download it, set up your crew, give it the first task — let the crew carry the technical work.
.dmg · macOS 13+ · Apple Silicon · v0.2.43
.exe · Windows 10/11 · 64-bit · 130 MB · v0.2.43
AppImage · Linux · x86_64 · 156 MB · v0.2.43
Intel Mac? Download the x64 build (157 MB)
Different platform?macOS (Apple Silicon)Windows 10/11Linux (AppImage)
- BYO-key — your own key
- Runs locally
- Cancel any time
Sign in with your Muratify ID after installing. No account? It takes a minute at accounts.muratify.com.
Sign in with Muratify ID