Routing tier
M-1 Lite
Fast and light on credits. Small tasks stay on the quickest model; only the heavier ones step up.
Best for Quick questions, rewrites, summaries
Chooses from, light to strong
- Qwen 3.7 FlashAlibaba Cloud
- GLM 5.3 FlashZ.ai
M-1
M-1 isn’t a model.
It chooses one.
ManyPilot doesn’t train its own AI. M-1 is our router: for every request it reads the task and hands it to the model best suited to it, from a pool built by other labs. Lite, M-1 and Pro are three pools — not three models.
Platform · built by ManyPilotManyPilot
The desktop app: pilots, skills, connectors, memory and the media studio.
Router · built by ManyPilot
M-1 LiteM-1M-1 Pro
Reads each request and picks a model from the tier’s pool. A tier is a choice of strength, not a model.
Models · built by other companies
Qwen 3.7 FlashAlibaba CloudGLM 5.3 FlashZ.aiDeepSeek V4 ProDeepSeekMiniMax M3MiniMaxKimi K2.6Moonshot AI
The AI that actually writes the answer. You’re billed for the one used, plus 15%.
Model routing
Watch M-1 choose.
Type a task, or pick one, and see how each tier scores it and which model it sends it to. Try the same task on Lite and on Pro.
M-1 Pro Never the lightest model: the hardest work reaches the strongest one.
1 Give it a task
2 M-1 reads the request
- Starting point25
- short request−10
- several demanding tasks+30
Demand45/100moderate
3 It picks the model
DeepSeek V4 Pro
MiniMax M3
Kimi K2.6
M-1 Pro sends this toMiniMax M3a model built by MiniMax, not by ManyPilot
Same task on M-1 Lite → Qwen 3.7 Flash, M-1 → GLM 5.3 Flash
This uses the same scoring rules as the live router. A real request also weighs how much context it carries, so scores in the app can sit a little higher. You’re billed for the model actually used.
The tiers
Three pools. You pick the strength.
Routing tier
M-1
The everyday all-rounder and the default. Covers everyday work, and reaches a stronger model when a task asks for it.
Best for Research, writing, most agent work
Chooses from, light to strong
- Qwen 3.7 FlashAlibaba Cloud
- GLM 5.3 FlashZ.ai
- DeepSeek V4 ProDeepSeek
Routing tier
M-1 Pro
Our strongest reasoning. Never the lightest model: the hardest work reaches the strongest one.
Best for Hard code, long tasks, deep analysis
Chooses from, light to strong
- DeepSeek V4 ProDeepSeek
- MiniMax M3MiniMax
- Kimi K2.6Moonshot AI
Works with the tools you already use
Your pilots reach into your apps.
Connect a tool once and choose which pilots may use it. Any MCP server works too.
Trello
Boards, lists and cards. Connect it with your Trello API key and token and a pilot can manage cards from chat.
Zapier
Run the Zapier actions you’ve switched on, across the apps Zapier connects. Sign in once.
Chrome
GitHub
Google Workspace
Microsoft 365
Notion
Slack
Linear
Jira & Confluence
Figma
Canva
Stripe
HubSpot
Supabase
Airtable
Asana
Todoist
Intercom
Sentry
GitLab
Hugging Face
YouTube
LinkedIn
Reddit
Telegram
Discord
Playwright
Logos are trademarks of their respective owners. Using a connector doesn’t imply endorsement.
How it compares
ManyPilot next to Zapier, Make and n8n.
They solve different problems. The automation platforms run workflows between apps in the background; ManyPilot gives you assistants that do the work with you. They also work together: ManyPilot has a Zapier connector, so a pilot can run your Zaps.
| Feature | ManyPilot | Zapier | Make | n8n |
|---|---|---|---|---|
| Built for | AI pilots that do the work: research, writing, code, browsing, media | Automating steps between cloud apps, with AI agents and an MCP server | Visual workflows (scenarios) that connect apps and AI | Node-based workflows, with AI agent nodes |
| Where it runs | Your Windows or macOS computer | Zapier’s cloud | Make’s cloud | Your own server, or n8n’s cloud |
| How you build | Describe the job in chat; give a pilot instructions, files and skills | Triggers and actions in an editor | Drag modules onto a canvas | Connect nodes on a canvas |
| Integrations | Ready-made connectors, including Trello, Slack, GitHub, Notion and Zapier, plus any MCP server | 9,000+ apps | Thousands of apps | 1,500+ integrations |
| Which AI model works | M-1 picks a model for every request. No API keys to manage | AI apps as steps; Zapier MCP lets the AI tool you already use run Zapier actions | AI provider modules as steps | You connect OpenAI, Anthropic, Google or local models yourself |
| Always on | Runs while the app and your computer are on | Yes, runs in the cloud | Yes, runs in the cloud | Yes, when hosted on a server |
| Your files and data | Chats, files and memory stay on your computer | Processed in Zapier’s cloud | Processed in Make’s cloud | Stay on your server when self-hosted |
| Best when | You want an assistant that works on your own files, browser and apps | You need triggers across many apps, running around the clock | You want to design branching, multi-step flows visually | You want to self-host and control every workflow |
Competitor details are taken from their public pages (Zapier MCP docs, Make integrations, n8n on GitHub) as of October 2026. Product names are trademarks of their owners; check their sites for current numbers.
Questions
M-1, in plain words.
Is M-1 an AI model?
No. M-1 is ManyPilot’s router. ManyPilot does not train its own foundation models. For every request, M-1 reads what was asked and sends it to a suitable model from the tier’s pool. The models themselves are built by other companies.
What is the difference between M-1 Lite, M-1 and M-1 Pro?
They are three pools of models with different strength. Lite sticks to fast, light models; M-1 adds a stronger one for everyday work; Pro never uses the lightest model and reaches its strongest one for the hardest tasks. You pick the tier, M-1 picks the model inside it.
Can I choose Claude, GPT or Gemini?
Not today. The tiers route only between the five models listed on this page. If that list changes, this page changes with it.
How is a request scored?
The router looks at the request itself: how long it is, whether it asks for demanding work such as debugging, research or design, whether it contains code, whether tools are switched on and how long the chat already is. The demo above uses the same rules. Once a chat steps up to a stronger model it doesn’t drop back, which keeps long conversations steady.
What do I pay?
Credits from your plan. A request costs what the model actually used costs, plus 15%, so a light task on a light model uses fewer credits. Every request shows in your dashboard.