Twin logo
Agents Orchestration Frameworks · Workflow Automation

Twin

Twin is a French-built platform for fully autonomous AI agents. Describe an outcome in plain English and Twin builds an agent that scrapes, browses, enriches and writes across your apps, then reports back in Slack on its own schedule.

Active GDPR compliant Free trial Subscription API available 16+ Verified by Guidaio
Overview

What is Twin?

Twin is a platform for building, running and distributing fully autonomous AI agents, published by Twin Labs SAS in Paris. Its central claim is that it is not a copilot: instead of waiting for a prompt, an agent triggers itself from a schedule, an inbound email, a phone call or a webhook, carries a whole process from request to result, and reports back in Slack. The company markets this as hiring an AI employee rather than adopting another chat window.

You describe the outcome you want in plain language and Twin designs the agent around it, choosing its own tools and steps. It reaches your stack in three ways. Native connections cover the common business tools, though the published counts are inconsistent: 2,700+ apps in the documentation, 5,000+ on the product page and 41,706 on the integrations page. Webhooks and a REST API cover custom systems and let you fire agents from your own backend. And where a tool has no API at all, Twin drives an embedded browser, logging in, clicking and extracting exactly as a person would. That third route is the real differentiator against workflow builders.

Nine capabilities sit in the box: web scraping, browser agent, slides, images and video, connection to any API, generated interfaces such as dashboards and CRM views, multichannel outreach, data enrichment, triggers and webhooks, and team sharing with a public marketplace. Under the hood, Twin separates an expensive Build Mode that uses high-reasoning models from a Run Mode three to ten times cheaper, and keeps long-term memory that an internal agent consolidates and prunes across hundreds of tasks. Everything runs in isolated cloud environments, so there is nothing to install or host.

A ready-made Employee Library covers prospecting, content, real estate lead hunting and buying-signal monitoring. Twin claims 85,000 users, 5,000 companies and three million agents built. It went public on 27 January 2026 alongside a $10M seed round, after a one-month beta in which more than 100,000 autonomous agents were deployed.

What it does

  • Build an autonomous agent by describing the outcome you want in plain language, with no code and no workflow nodes
  • Operate websites and web apps that have no API by driving a real browser: logging in, clicking, filling forms and extracting
  • Scrape structured data from any site, including pages behind authentication, and write it wherever you need it
  • Find, enrich and qualify leads with verified emails, phone numbers and addresses, then hand over a clean list
  • Run multichannel outreach across email, SMS, WhatsApp, LinkedIn and Facebook, with follow-ups sequenced automatically
  • Generate production-ready decks, images and video, and build dashboards, CRM views and workflow interfaces from your data
  • Trigger agents on a schedule, from an email, a call or a webhook, and get the result reported back in Slack
Audience

When to use Twin / When not to

A quick filter to help you decide if Twin is the right fit.

When to use Twin

  • Sales and growth teams that want prospecting, enrichment and multichannel outreach running every morning without anyone opening a tab
  • Operations and back-office teams stuck with internal tools, supplier portals or legacy dashboards that expose no API
  • Agencies and freelancers who want to build agents once, reuse them across clients and resell them on the marketplace
  • Non-technical founders and small teams who need a working automation in minutes rather than a workflow project
  • Developers who want to trigger agents from their own backend through the REST API instead of maintaining scrapers and connectors

When not to use Twin

  • Anyone looking for a conversational assistant to chat with: Twin is built to execute work, not to answer questions
  • Teams that need every step of a process pinned down in advance, node by node, with fully deterministic branching
  • Organisations that require a named security certification on the vendor's own pages, since Twin only links out to an external trust centre
  • Users who want a mobile app: the product is web-only, with a Slack app and an API and nothing for iOS or Android
  • Anyone needing a non-English interface, as no other language is announced for the product or its documentation
Get started

How to use Twin

A typical end-to-end flow, from setup to results.

  1. Create a free account on builder.twin.so — 1,000 credits are granted immediately and no credit card is required
  2. Pick a ready-made agent from the Employee Library, or open the chat and describe the automation you want in plain language
  3. Let Twin design the agent: it works out the steps, picks the tools and assembles the workflow around your goal
  4. Connect your stack — CRM, drive, spreadsheets, mailbox — through the native integrations; if a connector does not exist, Twin writes it
  5. For a tool with no API, authorise Twin to log in on your behalf so its browser agent can operate the site directly
  6. Choose a trigger: a recurring schedule, an inbound email, a phone call, a webhook or an event from a connected app
  7. Check the run-cost forecast the builder shows before you launch anything, so you know what the agent will consume
  8. Test the agent, then set it loose: it runs on its own and asks for explicit approval before any consequential action
  9. Collect the results where you work — Slack, email, your CRM or a generated dashboard
  10. Share the finished agent with your team, or publish it to the marketplace and charge for its use
Quick read

Pros & Cons

Pros

  • The embedded browser reaches internal tools, portals and legacy dashboards that classic workflow builders simply cannot touch
  • Genuinely no-code: an agent is described in plain language, and Twin writes the connector itself when none exists
  • Agents trigger themselves from a schedule, an email, a call or a webhook, so nothing depends on someone remembering to run them
  • Strong European posture: French publisher, customer data hosted primarily in the EU, subprocessor list published in full on the site
  • Written no-training commitments on Slack data and Google user data, backed by contracts with model providers
  • Cost is made visible: the builder forecasts run cost before launch, unused credits roll over, and idle agents are never billed
  • Low barrier to entry — 1,000 free credits without a card, a €20 entry plan, and email support with no account required

Cons

  • Credit-based pricing makes spend hard to predict, and Build Mode costs three to ten times more than Run Mode
  • No permanent free tier is published: the 1,000 credits are a one-off grant and all four listed plans are paid
  • The terms grant Twin a worldwide licence over de-identified Service Data to improve its services, lasting three years beyond the contract
  • No documented way for a customer to opt their data out of training; the no-training promises cover Slack and Google data only
  • Integration counts contradict each other across the site — 2,700+, 5,000+ and 41,706 depending on the page
  • No certification is named anywhere on the site: the "Audited & certified" badge points to an external trust centre
  • The terms date from May 2025 and still describe €0.05-per-step billing, which no longer matches the published credit plans
Pricing

Pricing & Plans

Twin does not publish a permanent free plan. New accounts receive a one-off grant of 1,000 credits with no credit card required, which functions as a free trial rather than a free tier. The lowest paid entry point is the Mini plan at EUR 20 per month, which includes 2,000 credits and roughly twenty agent tasks. All plans carry the same feature set; only the credit allowance, the number of agents and the level of support differ. Credits are consumed only when an agent actually works, and unused credits roll over to the next cycle.

Plan 1
  • Mini — EUR 20 per month — 2
  • 000 credits
  • around 20 tasks per month
  • 1 agent
  • manual top-up
Plan 3
  • Pro — EUR 189 per month — 20
  • 000 credits
  • around 200 tasks per month
  • up to 10 agents
  • automatic top-up
Plan 4
  • Max — EUR 882 per month — top tier shown in the plan selector
  • with no published credit breakdown
Plan 5
  • Enterprise — custom quote — shared credit wallets
  • teammate invites
  • per-member spending caps
  • private support and bespoke commercial terms
Special offers — 1,000 free credits granted on sign-up, with no credit card required · Unused credits roll over from one billing cycle to the next on every plan · Volume discounts of up to 11.8% on the largest one-off credit top-ups, valid for six months while the subscription stays active · An accelerator programme and a certification programme are advertised in the site footer
Prices and plans listed above may evolve. Always check the official pricing page before subscribing.
Trust & Privacy

Data, GDPR & hosting

A consolidated view of how Twin handles your data.

GDPR overview

Implementation is concrete and above average. Twin Labs SAS is a French company, so the GDPR applies directly and the CNIL is named as the reference authority. The privacy policy, last updated on 17 July 2026, sets out the legal bases it relies on: contract (Art. 6(1)(b)), legitimate interests (Art. 6(1)(f)), legal obligation (Art. 6(1)(c)) and consent (Art. 6(1)(a)). Access, rectification, erasure, restriction, objection, portability and withdrawal of consent are all listed, with a one-month response commitment and a separate 45-day track for CCPA/CPRA requests. Transfers outside the EEA rely on Standard Contractual Clauses, the UK Addendum or an adequacy decision. A DPA is available on request from the terms page. Everything goes through a single address, support@twin.so: no DPO is named, and no Article 27 representative is designated, which is expected for a company already established in the Union.

Who owns the data?

Twin Labs SAS is the data controller for the personal data it collects through the site and the product, and acts as your processor for the customer content its agents handle, under documented instructions and a Data Processing Agreement. Your prompts, files and process descriptions are treated as confidential customer data. The terms are more expansive than the privacy policy: by accepting them you grant Twin a worldwide, royalty-free, non-exclusive licence to access, reproduce, de-identify and otherwise use your Service Data to provide, support or improve its services, and this licence survives the contract by three further years. Feedback is licensed perpetually and irrevocably.

Reuse rights

Twin states it uses personal data to run and secure the product, authenticate users, bill, answer support requests and improve features from aggregated or de-identified usage. It says it never sells personal data and never uses customer content for advertising. Two explicit no-training commitments exist: Google user data is never used to train proprietary or third-party models, and Slack data is never used to train large language models, neither Twin's nor its providers'. Its agreements with model vendors such as OpenAI forbid them from improving their general models on your data. Nothing broader is promised, however, and the terms still allow de-identified Service Data to be reused to improve Twin or any other Twin service, including services offered to third parties. On the customer side, the outputs an agent produces are yours to use freely: the material is delivered into your own tools and the site imposes no reuse restriction on it. The published subprocessor list names Tailscale, Hetzner, Anthropic, OpenAI, AWS, Google Cloud, Stripe, WorkOS, Twilio and Axiom.

Data retention & training

Retention summary
Automation artefacts that contain your data — execution logs, screenshots, DOM traces, downloaded documents — are deleted within 30 days of a run, or sooner, with remaining traces removed from backups within a further 30 days. Disconnecting Google or Slack, or deleting your workspace, revokes tokens and clears cached data within 30 days. Account and workspace records are kept for the life of the workspace and deleted within 30 days of termination. Support communications and incident reports are kept up to three years, security and usage logs up to twelve months, encrypted backups up to 35 days, and invoices ten years under French accounting law. Shorter retention windows can be requested from support. Once a period expires, data is deleted or irreversibly anonymised.
Trains on customer data
Unclear
Subprocessors disclosed
Yes
DPA available
Yes
GDPR contact

Hosting summary

Twin states that customer data is hosted primarily in the European Union. Its published subprocessor list names Hetzner for server hosting in the EU, with AWS and Google Cloud handling EU backup storage and key management, Tailscale for internal networking and Axiom for log aggregation. Model inference runs on Anthropic and OpenAI, authentication and SSO on WorkOS, payments on Stripe and SMS verification on Twilio. Some of these providers, OpenAI among them, sit outside the EU, mainly in the United States; those transfers rely on the European Commission's Standard Contractual Clauses, the UK Addendum or an adequacy decision. Traffic is encrypted in transit with TLS 1.2 or higher and at rest with AES-256, credentials and tokens live in isolated secrets management with separate key rotation, and production access follows least privilege with mandatory multi-factor authentication and periodic review. No individual hosting country is named — only the EU region — and no security certification is stated on Twin's own pages, which instead link to an externally hosted trust centre.

Hosting regions
EU
Watch-outs

Things to keep in mind

Risks and trade-offs to weigh before adopting Twin.

  • Twin stores third-party credentials — passwords, one-time codes, API keys, browser session tokens — in its vault to operate sites without an API; the encryption is documented, but the concentration of access is a real exposure
  • The terms grant a worldwide licence over de-identified Service Data to improve Twin or any other Twin service, including services sold to third parties, for the contract term plus three years
  • No opt-out from training is documented anywhere: the existing promises are targeted at Slack and Google data, not a setting you control
  • Agents act autonomously across your tools, so a badly scoped instruction can send emails, update records or spend credits at scale before anyone notices
  • LLM output can be inaccurate or incomplete, and Twin says so; delegating a recurring process makes it easy to stop reading what the agent produces
  • Advertising trackers on the marketing site include RB2B, which identifies the company and, for some US visitors, the individual behind a visit
  • Usage and integration figures are vendor claims that nothing on the site substantiates: 85,000 users, 3 million agents built, 41,706 apps
Setup

Setup & Integrations

Technical difficulty

Low. There is nothing to install, host or configure: no server, no local machine, no key management. You create a free account, receive 1,000 credits without a card, and describe the automation you want in a chat. Twin designs the agent, and writes the connector itself when a tool has no integration. Connecting apps is an OAuth flow. The vendor targets non-technical users explicitly and claims a first agent in fifteen seconds. The residual effort is judgement rather than engineering: scoping credit consumption, and deciding how far to let Twin store third-party credentials for sites with no API.

Deployment

Web appAPISlack app

Integrations

Google Sheets Google Docs Google Drive Gmail Google Calendar Google Analytics Google Search Console Google Ads Google Gemini Notion Airtable Calendly Typeform ClickUp Slack Telegram Discord Twilio Resend Mailgun Buffer HubSpot GoHighLevel Apollo Follow Up Boss Attio Clay Salesforce Stripe Shopify Etsy Amazon EBay AliExpress Gumroad GitHub Supabase Firebase Replit Railway Firecrawl Apify Jina AI Perplexity ElevenLabs HeyGen Shotstack Gamma LinkedIn X (Twitter) Reddit TikTok YouTube Instagram Facebook WhatsApp Substack Skool Meta Ads Microsoft Graph Microsoft 365 Outlook Wix Zillow Redfin Realtor.com LoopNet Indeed Vapi Retell AI Base44 CoinGecko Coinbase Binance Kalshi Yahoo Finance Finviz Alpaca SEC EDGAR Open Meteo TechCrunch Reuters The Verge VentureBeat Hacker News ESPN SAM.gov Qonto PostgreSQL Excel Microsoft Teams Dialpad
Company

Behind Twin

Company name
Twin Labs SAS
Founded
02/08/2022
Country of origin
🇫🇷 France
Headquarters
10 rue de Penthièvre, 75008, Paris, France
UBO
Hugo Mercier
UBO country
🇫🇷 France
Domain registrar country
🇺🇸 United States
Support contact

Fundraising

January 2026 — $10M seed round led by LocalGlobe, announced by Twin on the day of its public launch
February 2024 — $3M pre-seed round led by Betaworks, with Motier Ventures and Factorial alongside a group of business angels (reported by third-party press, not by the site itself)

Social

Official links

Resources

All the official URLs gathered for verification and reference.

Compare

Alternatives

Tools that compete with or complement Twin.

Z ZapierM MakeN n8nD DustL LindyG GumloopM Microsoft Copilot StudioC ChatGPT AgentC Claude CoworkC ChatGPTV Viktor
FAQ

Frequently asked questions

How does Twin's pricing actually work?
Twin runs on credits. They are deducted only when an agent does real work — building, running, browsing, researching or generating output — and never for idle agents, saved configurations or simply viewing results. Each monthly plan comes with an allowance, and unused credits roll over into the next cycle.
What does a typical task cost in credits?
Twin publishes rough orders of magnitude: 15 to 30 credits for a simple automation, 20 to 70 for a scraping job of around 100 items, 100 to 200 for a browser-agent session of roughly 20 steps, and 20 to 50 per generated image, video or deck. Building an agent is three to ten times more expensive than running it.
Is there a free plan or a free trial?
There is no permanent free tier. Signing up grants 1,000 credits immediately with no credit card required, which lets you build and run a first agent before paying. Beyond that, the cheapest published plan is Mini at EUR 20 per month.
What happens if the tool I need has no API?
That is Twin's core strength. It drives a real embedded browser: it logs in, navigates the interface, reads pages, fills forms and completes actions the way a person would. This covers internal portals, supplier dashboards and directories that traditional automation tools cannot reach.
Do I need to know how to code?
No. You describe the outcome you want in plain language and Twin designs the agent, picks the tools and assembles the steps. If a connector does not exist, Twin writes it. Developers who want a different route can trigger agents from their own backend through the REST API.
Where is my data hosted?
Customer data is hosted primarily in the European Union. The published subprocessor list names Hetzner for server hosting in the EU, with AWS and Google Cloud for EU backup storage and key management. Some providers, including OpenAI, sit outside the EU, and those transfers rely on Standard Contractual Clauses or an adequacy decision.
Does Twin train its models on my data?
Twin commits in writing not to train large language models on Slack data, neither its own nor its providers', and not to use Google user data to train proprietary or third-party models. No broader commitment exists, and the terms still allow de-identified Service Data to be reused to improve Twin's services, so the overall stance is not fully settled.
Is there a DPA, and is there a minimum age?
A Data Processing Agreement is available on request through the link at the bottom of the terms page. The services are designed for business users and are stated as not intended for children under 16.
Is there a mobile app?
No. Twin is a cloud product used from the web builder, with an official Slack app that brings agents into your channels and direct messages, plus a REST API. Nothing is published for iOS or Android.
How do I reach support?
By email at support@twin.so, with no account or sign-up required. Twin says it answers every support request within two business days and usually faster. In-app chat and full documentation are available on every plan.
Conclusion

Should you pick Twin?

Twin occupies a genuinely distinct position. Where most automation platforms ask you to wire triggers and map fields, and where most AI assistants stop at giving advice, Twin sets out to run the whole process and to start it itself. The embedded browser is what makes that credible: an agent that can log into a supplier portal or an internal dashboard reaches the work that no connector-based tool ever will.

For European buyers the compliance posture is unusually solid for a company this young. A French publisher, customer data hosted primarily in the EU, a subprocessor list published in plain sight rather than behind a form, a DPA on request, and written no-training commitments on Slack and Google data all count for something.

The reservations are real and worth weighing. Credit pricing makes spend genuinely hard to forecast, and the gap between an expensive build phase and a cheap run phase means early experimentation costs more than steady operation. There is no permanent free tier despite the "start free" framing. No certification is named anywhere on Twin's own pages. And the terms, still dated May 2025, both grant a broad licence over de-identified customer data for three years beyond the contract and describe a per-step billing model that the current pricing page has abandoned — a contract that has not kept pace with the product.

Twin is worth a serious trial for teams whose bottleneck is genuinely operational: repetitive work spread across tools that refuse to talk to each other. Start on the free credits, build one agent that matters, and watch what it consumes before committing. Teams needing certified compliance evidence or contractual clarity on data reuse should get answers in writing first.