Twin Labs

Twin Labs SAS, operating the product Twin at twin.so, develops an AI agent platform that lets users create and deploy autonomous "AI employees" to run recurring business processes. The product is presented as a library of ready-made agents — for example roles described as Briefing Analyst, Lead Scout, Outreach Operator, Marketing Producer and Operations Agent — alongside a builder for custom agents assembled from capabilities including web scraping, browser control of web apps without APIs, generation of slides, images and video, custom dashboards and workflow interfaces, multi-channel outreach (email, SMS, WhatsApp, LinkedIn, Facebook), data enrichment, and triggers/webhooks. Twin advertises more than 5,000 native integrations and states that agents run on a schedule or are triggered by an email, call or webhook, reporting results in tools such as Slack and Gmail.

Technically, the company describes Twin as a multi-agent system in which an orchestrator routes intent, a builder creates the automation (API integrations, browser actions, code execution) using high-reasoning models, and a runner executes it autonomously using smaller, cheaper models; agents are described as self-healing when integrations break by diagnosing the failure, searching documentation, patching the tool and memorizing the fix. The stack is built in Rust and Nix, with a custom headless Wayland compositor ("waysmoother") that combines composition, H.264 encoding and WebSocket streaming for computer-use agents, and NixOS-based deployments on Hetzner Cloud where each session receives a sandboxed compositor and Chromium browser process. The company states its browser-use agent underpins roughly 80% of user workflows.

At launch the approach was based on multimodal vision models such as GPT-4V to replicate human actions in a browser — loading pages, clicking buttons and entering text — rather than relying on API-based, multi-step workflow builders. The initial plan was to ship a library of pre-trained tasks first and later open the platform so customers could create their own tasks.

Founding story

Twin Labs was founded in 2023 by Hugo Mercier (CEO) and Joao Justi, who spent roughly six months building a prototype before the company's pre-seed announcement in January 2024. Mercier described the starting point as testing whether an AI agent could be trained to duplicate the way people perform their tasks; after finding traditional LLM-based autonomous agents unreliable, the team moved to multimodal vision models, which could interpret software interfaces and the features behind on-screen buttons.

Business model

Twin sells self-service subscriptions to its agent platform, sized by usage credits, with the same feature set across tiers. Published plans are Mini at €20/month, Mini+ at €50/month for 5,000 credits (approximately 50 tasks/month), Pro at €189/month and Max at €882/month, with credit rollover, API connectivity, browser agent and web scraper, and native integrations included; a free tier is offered as a starting point. The terms of service describe monthly contracts that renew automatically each month unless terminated with 15 days' written notice, and provision of services via the Twin API and a web "Playground" application including browser infrastructure and model inference.

Recurring monthly subscriptions priced by credit consumption tied to the number and frequency of agent tasks, from a free entry tier up to higher-usage plans (€20 to €882 per month), with usage forecasting shown in the agent builder before launch.

Traction

The company's site reports 85,000 people using Twin across 5,000 companies, 3 million agents built, and a rating above 4.5 from more than 300 reviews, along with usage counts for popular agent jobs (for example 2,722 hires for prospecting and 1,373 for automated posting). Its GitHub organization README cites 1,000,000 workflows automated, 100,000 agents deployed and 60,000 users.

Latest developments

Terms of service covering the Playground, API, browser infrastructure and model inference were last updated in May 2025. The GitHub organization README reports cumulative totals of 1,000,000 workflows automated, 100,000 agents deployed and 60,000 users, cites €12M raised from LocalGlobe and founder angels, and links to open positions; organization repository activity is recorded into 2025 and 2026.