Muse 01
Tested on hundreds of simulations
Each scenario is one YAML file. Simulated users try every edge case before a real one does.
Build once.
Ship to every chat.
Let users chat with your product on iMessage & WhatsApp, and use it as an MCP in their ChatGPT, Claude.
Start building
Copied. Give it to your coding agent (Claude Code, Codex, Cursor) and it will do the full setup.
Backed by 
Entrepreneurs First
Illustration: DoorDash, HubSpot and the other brands on this page are examples, not Flow customers, and the numbers are made up.
Ships to
- iMessage
- Claude
- ChatGPT
- MCP
- CLI
For a decade, people used apps.
Dashboards at work. Mobile apps for everything else. Click, tap, fill in the form.
Now they'd rather just chat.
Say what they want, the way they'd text a friend, and get it done right there.
And they chat everywhere.
ChatGPT and Claude. WhatsApp and iMessage. Personal agents like Instinct and Muse. Any MCP client. A terminal.
Flow puts your product in every chat.
One agent, one codebase, every place your users ask.
Start with our open-source framework, built for the interfaces of the future.
Ready for production
Ship it like software.
Tested, traced and improving on its own.
02
Self-improving MCPs and agents
When the model misreads intent or a tool is unclear, Flow suggests a fix.
03
Traced, every turn
Every model and tool call, timed and kept, so you see why it said what it said.
Illustration: the numbers are examples, not real results.
For consumer apps
WhatsApp and iMessage agents
that feel like texting back.
01
AI orchestrator
A human-like chat experience. It handles shorthand, second thoughts and two languages in one sentence, then does the thing.
02
Built-in memory
It remembers each customer's chat between conversations, on each channel.
03
Voice notes, photos and files
Customers talk, snap and send the way they already do. It understands all of it.
04
Reaches out at the right time
Reminders and follow-ups in each person's time zone, never during quiet hours.
For SaaS
MCPs and CLIs
for ChatGPT, Claude and coding agents.
01
Your product inside Claude and ChatGPT
One MCP server from the same codebase, so people use your product where they already work.
02
A CLI for coding agents
Claude Code and Codex drive your product from the terminal.
03
Each user signs in
OAuth per user, so the agent only sees what they can.
Illustration: HubSpot and the other brands are examples, not Flow customers.
Pricing
Free to start. Pay per user, not per message.
A user is anyone who uses your agent, through an MCP client, a CLI, ChatGPT, Claude or a chat app. No per-message fees on any plan; iMessage plans cap different people a day and a month (below).
Starter
$0 forever
Launch to real users in minutes.
Get started- Up to 25 users
- MCP, CLI, ChatGPT and Claude
- Shared iMessage and WhatsApp line
- Hosting, Test Suite and Analytics
Pro
Most popular$25$38$68$118$193$343$643 /mo
Scale as your users grow.
Get started- 500 users included
- $10 of agent credit every month
- MCP, CLI, ChatGPT and Claude
- Hosting, Test Suite and Analytics
Enterprise
Custom
Our team helps you set it up and launch.
Book a call- Hands-on setup with our team
- Custom pricing and volumes
- Compliance and security review
- Every surface and channel
WhatsApp fees are billed by Meta directly, with no markup.
iMessage is priced on its own: a free trial on the shared line for 25 people, $25 a month for the shared line (300 people a month, 200 a day), $100 a month for your own number (1,000 people a day with no monthly cap; a new line starts at 250 a day and ramps to 1,000 over its first 14 days), and $15 a month more for SMS/RCS fallback and call forwarding on your own number. Messaging people first (people who haven't written to the line yet) is $40 a month more with Outbound, on your own number. Monthly, no setup fee; buy them in the dashboard. iMessage plans and limits
In every plan
- No per-message fees
- Agent usage at cost, no markup
- Every conversation kept and traced
- No lock-in, your data exports
Who we are
Two people who want your app everywhere your customers chat.
We build Flow ourselves, and we answer every message ourselves. Say hi on LinkedIn or book a call.
Book a call
FAQ Questions,
Questions,
answered.
What is a WhatsApp or Telegram AI agent?
An AI agent lets customers use your product through a conversation. With Flow Framework you write your product’s jobs as typed tools that call your APIs, so the agent can answer questions and take actions, such as checking an order or booking an appointment, on iMessage, WhatsApp or Telegram.
Do I need to have an app already?
No. The agent’s tools call whatever you already run: an API, a database or an MCP server. Until those are ready, a tool can return sample data, so you can text the agent on the Telegram sandbox the same day and swap in real calls later.
Do I need engineers to set it up?
Someone who can write TypeScript, or a coding agent. Copy the agent instructions at the top of this page into Claude Code, Codex or Cursor and it creates the project, writes the tools and runs the tests. Real data, signed-in actions and payments are usually an afternoon for whoever owns your backend.
How does it handle orders and payments?
Actions run against your real systems, never guessed, and every one is logged so you can audit it. Before a release goes out, Flow runs it against hundreds of simulated users on the paths that matter, ordering, rescheduling, refunds, and people changing their mind mid-sentence.
What about platform approvals?
Every surface has its own review and its own rules, such as WhatsApp's message templates and the ChatGPT and Claude app directories. You submit to each platform under your own account, and the timelines belong to the platforms, so we don't promise them.
Does it remember a customer between conversations?
Yes. On each channel, the agent keeps the person's conversation there, so they can pick up where they left off. You can delete everything kept about a person at any time.
Who owns the conversations and the data?
You do. The threads, the contacts and the orders are yours and export at any time. We do not sell data and we do not train public models on your customers.
How do agent credits work?
Pro comes with $10 of agent credit every month to spend on LLM tokens: your agent's WhatsApp, iMessage and CLI chats, and any LLM calls in your MCPs. Tokens past the $10 are billed at cost, with no markup. Unused credit doesn't roll over.
Which AI models does my agent use?
You get direct access to a few models we have proven for chat, so your agent works well from day one without choosing or setting up a model.
Can I bring my own model?
Yes. The Flow Framework lets you plug in your own model and provider, so you're never locked in. When your agent runs on your own model, its usage is billed by your provider, not by us.
How is Pro priced as I grow?
Pro is $25 a month with 500 users included. Each user after that costs less the more you have: $0.025 up to 1,000 users, $0.02 up to 5,000, $0.015 up to 20,000, and $0.01 after that. A user is anyone who uses your agent in a month, through any surface.
How do I get started?
Copy the agent instructions at the top of this page into your coding agent, or run npx @flow-engineer/flow new my-agent. Then sign up at flow.engineer for a free trial of 25 people per channel on Flow’s shared lines. No card and no call needed.
The messaging apps around it: WhatsApp, iMessage and Telegram.
Launch your app
where people already chat.
Copy the agent instructions and your coding agent does the setup. No card, no demo call.