Up to 32 agents in parallel
One goalMany minds.
SyntobiMind runs Claude Code, Codex, Antigravity and Cursor agents in parallel on your project. Simorgh plans the work, up to thirty-two agents ship it — and nothing lands until you approve.
Simorgh, the conductor
Watch one round, from goal to merge.
Type one goal. Simorgh plans it, a mixed crew of AI coding agents builds it in parallel, every branch is reviewed — and nothing reaches your project until you merge.
Simorgh’s plan · 7 parallel agents
Simorgh reads your project and writes the plan
One sentence in, English or Persian. Simorgh maps the codebase, writes a shared blueprint and one brief per task, and gives each agent its own files — so no two agents touch the same code.
Spend less, ship more
Your top model thinks. Lighter models build.
Simorgh spends your strongest model only where judgement matters — the plan, the briefs and the review — and hands the implementation to faster, lighter models that follow those briefs. Architect-quality work, without burning the top model on every line.
Top-model share of the work100%
All of itAbout one fifth
5-hour usage limit100%
Limit hit mid-round — wait for the resetRoom left for more rounds
Time to done100%
Slower: the big model types every lineFaster: light models type quicker, in parallel
Illustrative example of one 8-task round, not a benchmark. Real usage depends on your plans, models and project.
How does SyntobiMind lower AI model usage?
SyntobiMind’s conductor, Simorgh, uses a strong model only as the architect: it plans the work, writes a precise brief for every task and reviews every branch. Faster, cheaper models implement the tasks in parallel from those briefs. Most tokens are spent by the lighter models, so the top model’s 5-hour and weekly limits — Claude’s, for example — last much longer, and rounds finish sooner.
Four views, one project
One app. Every way to work.
Switch views without leaving the project — the same folders, the same agents, the same round.
Every agent of the round in its own live pane — watch the whole crew at once.
Up to 32 panesSteer mid-runPlan tab beside
Made for you
Built for vibe coders. Trusted by pros.
Say what you want. Watch it get built.
Plain words in, working app out
Describe the goal in English or Persian. Simorgh plans it, a crew builds it, the browser opens the result.
No git, no terminal
Branches, merges and backups happen for you. If you don’t like a change, Undo puts everything back.
Safe by default
Agents ask before risky commands. You click Allow or Deny — nothing touches files outside your project.
Your subscriptions, no surprise bills
Runs on the Claude, ChatGPT, Google and Cursor plans you already have.
Build a booking site for my barbershop — with an admin panel and SMS reminders.
Simorgh’s plan
- ✓Booking page & calendaragent 1
- ✓Admin panel & sign-inagent 2
- ✓SMS remindersagent 3
- ✓Tests & a live previewagent 4
Done — the preview is open in the built-in browser.
Scale
From one agent to thirty-two.
1
agentStart with one.
Ask in plain words — English or Persian. One agent reads the project and gets to work.
SyntobiMind
AgentCodeThreadChat
—▢✕Round 11 agent
Every provider
Claude Code, Codex, Antigravity and Cursor — side by side.
SyntobiMind runs the official coding CLIs on the subscriptions you already pay for. Mix them in one round, each agent on the model and effort its task needs.
Claude Code
Claude Pro / Max
● Read src/checkout/Checkout.tsx
● Edit 3 files +88 −12
● npm test ✓ 44 passed
Opus · Sonnetworking
Codex
ChatGPT plan
● apply_patch src/cart/api.ts
● vitest run src/cart
● ✓ 12 files green
GPT · Codexworking
Antigravity
Google AI plan
● Plan: search index
● Write src/search/index.ts
● Check: 9 tests pass
Geminiworking
Cursor
Cursor plan
● Read README.md
● Edit src/theme/tokens.ts
● Done · branch ready
Composer · Autoworking
Sign in once inside each official CLI. Your accounts, your limits — no middleman, no extra API bill.
Bring any model
Plug in any API — even the models on your own PC.
Paste a key from OpenRouter, DeepSeek, Groq, Together, Mistral, Gemini, Anthropic, OpenAI or xAI — or the base URL of any OpenAI-compatible server, including Ollama, LM Studio, vLLM and LiteLLM. Its models show up in the picker and can join the team as agents or answer in Chat.
Try it — pick one
/models
Use it as
Arash…
openrouter:anthropic/claude-sonnet-4.5
Code-mode agent · the app runs its file and command tools
Example model names; the real list comes from your provider.
Three API formats
OpenAI Chat Completions, OpenAI Responses and Anthropic Messages — what nearly every provider and gateway speaks.
Local models, no key
Point it at Ollama, LM Studio or a llama.cpp/vLLM server on your PC; no key, no bill.
Spending caps
A daily cap and per-run and per-agent budgets; the cost of every message is shown.
Keys stay yours
Encrypted with Windows DPAPI, stored only on this PC, never sent to SyntobiMind.
Between rounds
Tasks, memory and your servers, kept per project.
Everything a crew needs to pick up where it left off — and the remote machines it works on.
A to-do list your agents can read and write
Plan the week for each project. Agents suggest new tasks and check off the ones they finish; hand any task to an agent with one click.
- Week view with due dates and flags
- Agent suggestions you keep or dismiss
- “Work on all” sends the list to Simorgh
Control room
You stay in control.
Steer mid-run, approve what lands, and keep every agent on the same rules — without leaving the desk.
4 agents · 1 needs you
01
agent · auth
● Edit src/auth.rs
⎿ 14 tests passed
Live steering
Chip a note mid-run — Persian or code. The next step reads it.
02
Every write waits for you.
Allow?
cargo test
Approval
Allow or Deny a real queue. Dangerous shells get a red stamp.
03
Worktrees & Undo
Branches fork from main. Merge, then rewind with one click.
04
Your subscriptions. No API bills.
Usage rings
Hover a ring for 5-hour and weekly windows.
05
A1A2A3A4
Every agent inherits it
Team access
Add a rule once. Every agent inherits it.
06
:5173 → forwarded
Remote & Preview
SSH stays alive. Open the forwarded preview in place.
Persian, first-class
Persian, rendered right.
Terminals break Persian: letters fall apart and sentences run backwards. SyntobiMind draws every answer itself, so your agents read and write Persian as naturally as English.
Rendered by the app
Joined letters, correct order, Persian fonts bundled — in every mode.
Direction per paragraph
Persian, English and code each keep their own direction in the same answer.
Goals in Persian
Write «با هشت تا ایجنت» in your goal and Simorgh launches eight.
Persian dates
In-app announcements show Jalali dates for Persian text.
Rostam · Claude Code
فایل auth.rs را بازنویسی کردم و ۱۴ تست با موفقیت پاس شد. برای اجرای دوباره:
cargo test --all
اگر بخواهید ورود با گوگل را هم اضافه میکنم.
A real capture
Persian paragraphs, an English table, code — each in its own direction.
An agent reporting in Persian inside the app: joined letters, right-to-left paragraphs, Persian digits in the table, and code that stays left-to-right.
Everything inside
Everything a crew needs. Already inside.
now · 9 files changed#41
- +src/auth.rs
- +src/routes/login.ts
- +tests/auth.spec.ts
- +README.md
- +src/db/schema.sql
- +web/Login.tsx
Checkpoints & Rewind
Every turn is a checkpoint. Drag back to any moment — files and all.
Built-in browser
Agents open what they build right next to the code. Links from any pane land here too.
Mon
Tue
Wed
Thu
Fri
Sat
Sun
click a task to hand it to an agent
Tasks & Memory
A week runway per project. Hand any task to an agent; memory briefs every turn.
Command palette
Ctrl+K: modes, workspaces, agents, chats and settings in one box.
MCP servers, one page
See and change the MCP servers of Claude Code, Codex, Cursor and Antigravity in one place.
Sessions & lineups
Ctrl+N: pick a lineup — providers, models, effort — and launch side by side.
Chat & Compare
Nine API vendors, per-message cost, two or three models answering at once.
Lite mode & themes
Dark or light. Lite drops blur and motion for older PCs.
Signed updates
Updates are verified against SyntobiMind’s signature before they install.
0parallel agents
0agent CLIs + 9 API vendors
0API bills for agents
0%native Windows · Rust
Local-first
Your code never leaves your machine.
SyntobiMind is a native app, not a cloud service. Agents run on your computer through the official tools you sign in to — we never see your files, prompts or answers.
Runs where your code is
Agents work in your folders, with your CLIs and your accounts. No proxy, no relay, no upload.
Keys stay encrypted
Optional API keys are sealed with Windows DPAPI and never handed to agents.
Signed, verified updates
Each installer is signed by our license server and checked against its SHA-256 before it runs.
Nothing left running
Every agent stops with the app. Approval cards gate anything that reaches outside the project.
Pricing
Pick your crew.
Every plan runs on the AI subscriptions you already have — you pay for SyntobiMind, never for tokens.
Save up to 20%
Starter
Up to 3 independent agents side by side — No Orchestration.
1,000,000Toman/mo
- Up to 3 parallel agents
- Simorgh orchestration
- Claude Code · Codex · Antigravity · Cursor
- Chat mode & model Compare
- SSH remote servers
- 1 device
- Signed updates & support
Pro
Simorgh orchestration, Chat and SSH Servers on every CLI Provider.
Recommended
1,500,000Toman/mo
- Up to 16 parallel agents
- Simorgh orchestration · teams up to 8
- Claude Code · Codex · Antigravity · Cursor
- Chat mode & model Compare
- SSH remote servers
- 1 device
- Signed updates & support
Max
Everything in SyntobiMind — every Provider including your own API keys.
2,000,000Toman/mo
- Unlimited parallel agents
- Simorgh orchestration
- Claude Code · Codex · Antigravity · Cursor
- Agents on API models
- Chat mode & model Compare
- SSH remote servers
- 3 devices
- Signed updates & support
How buying works
- 1
Download & sign up
Install SyntobiMind and create your account in the app.
- 2
Pick a plan
Right after sign-up, or later from Settings › Account. Upgrades are prorated to the day.
- 3
Card-to-card, auto-confirmed
Transfer the exact amount shown — it’s matched and your plan turns on within moments.
A native Windows app that runs AI coding agents — Claude Code, Codex, Antigravity and Cursor — side by side on your project folders. Simorgh, the conductor, turns one goal into independent tasks, gives each to its own agent in its own git worktree, reviews what comes back and merges it.
Both. Describe what you want in plain English or Persian and Simorgh plans and ships it — no git or terminal knowledge needed. Professionals get the controls they expect: worktrees, per-task models and effort, approval cards, team rules, checkpoints, SSH servers and a review desk before anything lands.
No. Agents run on the Claude, ChatGPT (Codex), Google (Antigravity) and Cursor subscriptions you already pay for, through their official command-line tools. API keys are optional and only used by Chat mode.
Less than you might think. Simorgh uses your strongest model only as the architect — planning, writing a brief for every task and reviewing every branch — while faster, lighter models do the implementation from those briefs. Most of the work runs on the lighter models, so the top model’s 5-hour and weekly limits last much longer.
Each agent works in its own git worktree, told the scopes of the others. The review desk checks every branch before it merges, merges keep a backup with one-click Undo, and checkpoints let you rewind any turn.
No. Agents run on your computer and talk only to the AI provider you signed in to. Our license server sees your account, plan and device — never your files, prompts or answers.
Yes. SyntobiMind renders every answer itself, so right-to-left text, joined letters and mixed Persian and code read perfectly — something terminals get wrong. Write goals in Persian; the agents answer in it.
Windows 10 or 11 (64-bit) with the Microsoft Edge WebView2 runtime (the installer checks it), plus the agent tools you want to use. SyntobiMind finds them on your PATH and helps you connect each account.
Not yet. SyntobiMind is available for Windows today; macOS and Linux versions are in development. They will be announced on this site and inside the app when they are ready.
Choose a plan inside the app. You pay by card-to-card transfer; the payment is matched and confirmed automatically, usually within a minute, and your plan switches on at once — no restart, no receipt to upload. Plans don’t renew by themselves.
Yes. Add an SSH server and agents run there in sessions that survive network drops and closing the app, with ports forwarded so previews open in the built-in browser.
Ready to conduct your own crew?
Native on Windows, built around the AI subscriptions you already own, and fluent in Persian. Install in a minute — your first round is one sentence away.
- Version
- v0.44.2
- Size
- 15.6 MB
- Released
- October 31, 2026
In-app updates are signed and verified before they install.
Verify the download
The installer’s SHA-256 checksum. Optional: compare it with the file you downloaded (PowerShell: Get-FileHash).
SHA-256
25ab923bbf123d5886617ea454be994afcd6257bdd8018850dc32ab3364791cf