H

Automation & Agents , Productivity , Developer Tools

A specialized 'Action' model ecosystem that prioritizes UI interaction over conversational fluency, making it a high-performance choice for developers building autonomous web agents.

Excellent for high-volume browser automation and UI-based workflows, weaker for general-purpose chat or creative writing.

Ai Agents Automation Browser Automation Enterprise Api Workflow Automation Autonomous Agents Web Automation

Analysis based on product data, pricing structure, traffic signals, and public user sentiment.

Visit Website

Share this tool

Facebook X

LinkedIn Reddit

WhatsApp Telegram

Who Should Use H?

Typical users

AI engineers, QA automation teams, and enterprise developers building custom agentic workflows.

Maturity fit

advanced

Choose this if…

You need an agent that interacts with UIs via vision rather than just DOM parsing
Your priority is low-latency execution of multi-step digital tasks
You want to reduce token costs compared to using GPT-4V or Claude for browser use

Skip this if…

You need a polished, consumer-ready 'AI employee' with a simple GUI
Your workflow requires 100% reliability without human-in-the-loop oversight
You lack the technical resources to integrate via API or manage agent orchestration

About H

H is a Paris-based AI lab founded by former DeepMind and Mistral researchers to develop 'Action-Oriented' foundation models. Unlike standard LLMs, H's models are specifically trained to navigate digital interfaces, perform clicks, and execute multi-step workflows across websites and applications.

Official profiles

X (Twitter) LinkedIn

What it actually does

The platform provides a suite of models and tools, including Runner H and Surfer H, that interpret screen visuals to perform tasks like data entry, web scraping, and software testing. It translates natural language instructions into a sequence of mouse movements, keystrokes, and navigation actions.

What makes it different

H uses proprietary 'Hollow One' vision-language models (VLMs) that are significantly smaller and faster than general-purpose models like GPT-4. By focusing on UI-specific coordinates and action sequences rather than broad knowledge, it achieves higher accuracy on web-navigation benchmarks at a fraction of the inference cost.

Visual UI interpretation and coordinate prediction Autonomous multi-step task planning Parallel agent execution for high-volume tasks Self-healing automation that adapts to UI changes Native integration with Slack, Notion, and Google Workspace API-driven browser automation via Surfer H Human-in-the-loop validation for critical steps

Key Features

Runner H

The primary agentic model for executing complex, multi-app workflows.

Surfer H

An open-source framework for browser-based automation using vision.

Hollow One Models

Compact 3B and 7B VLMs optimized for UI interaction.

H-Studio

A visual environment for building, debugging, and monitoring agent runs.

WebClick Dataset

A massive open-source dataset for training click-level accuracy.

Parallel Fleet Deployment

Ability to spin up multiple agents to handle tasks simultaneously.

Cross-App Orchestration

Moves data between web browsers and productivity tools like Notion or Slack.

Pricing

Free Beta

Free

10 free runs for testing
Access to Runner H web interface
Basic Surfer H browser agent
Standard community support

Popular

Pro (Upcoming)

Usage-based per task

Full API access
Higher concurrency limits
Access to 7B Hollow One models
Priority execution

Enterprise

Custom annual

Private model deployment
SLA-backed support
Custom fine-tuning on internal UIs
Dedicated account management

Pricing checked 4 months ago

Pricing guidance

Best plan for most users: The Pro plan is the target for most developers, as it provides the API access necessary to build H into existing applications.

Free plan enough? No — the 10 free runs are strictly for a 'vibe check' and are insufficient for building or testing a real workflow.

Upgrade when:

When you need to automate more than 10 tasks
When you require API access for programmatic execution
When you need the higher accuracy of the 7B model

Watch out for:

Rate limits on the free beta are aggressive
Token costs for long-running browser sessions can add up quickly
Vision-based agents consume more bandwidth than text-only models

Premium technical positioning with a focus on cost-per-task efficiency for developers.

Pros & Cons

Strengths

Cost-effective action execution
The 3B Hollow One model can perform tasks for approximately $0.13 per run, significantly cheaper than using flagship models for the same UI-heavy tasks.
Vision-first navigation
Because it 'sees' the screen rather than relying on underlying code, it can navigate complex SPAs and dynamic websites that break traditional scraping scripts.
High-caliber technical foundation
Built by the team behind AlphaGo and Mistral, the architecture is designed for reasoning and planning rather than just text prediction.

Weaknesses

Early-stage reliability issues
Agents can occasionally get stuck in navigation loops or lose context when moving between multiple browser tabs.

Affects: Production-grade enterprise workflows
Opaque pricing for scale
While a free tier exists for testing, the long-term cost structure for high-volume enterprise API use remains unlisted.

Affects: Budget planning for scaling startups
Steep learning curve for orchestration
Setting up complex 'fleets' of agents requires a solid understanding of prompt engineering and agentic logic.

Affects: Non-technical business users

Real User Sentiment

Users are impressed by the vision-to-action speed but remain cautious about the 'agentic' promise of 100% autonomy.

Users tend to like

Speed of the Hollow One models
Ability to handle websites without APIs
The pedigree of the founding team
Open-source availability of the Surfer H framework

Users commonly complain about

Agents getting 'stuck' on simple UI hurdles
Lack of documentation for advanced API features
Limited availability during the private beta phase

Recurring tradeoffs

Lower cost vs. lower reasoning depth compared to GPT-4o
Vision-based flexibility vs. the fragility of UI changes

Happiest users

Developers building niche automation tools for sites that actively block traditional scrapers.

Often frustrated

Business operators expecting a 'set it and forget it' virtual assistant that never makes mistakes.

Use Cases

Dynamic Web Scraping

Extracting data from company career pages that use complex JavaScript.

QA Regression Testing

Running end-to-end visual tests on web apps using natural language scripts.

Lead Generation

Navigating LinkedIn or industry directories to compile prospect lists into Google Sheets.

Form Automation

Filling out repetitive government or enterprise forms that lack an API.

E-commerce Monitoring

Tracking price changes or stock levels on sites with aggressive bot detection.

Frequently Asked Questions

How much does H cost?

During the current beta phase, users get 10 free runs. H is transitioning to a usage-based model where tasks are expected to cost between $0.10 and $0.30 depending on the model (3B vs 7B) and task complexity.

How does H compare to Anthropic's Computer Use?

Anthropic's Computer Use is a feature of their flagship Claude models, which are larger and more expensive. H uses smaller, specialized models (Hollow One) that are faster and cheaper for high-volume UI tasks, though Claude may have better general reasoning for highly abstract instructions.

Is Surfer H open source?

Yes, the Surfer H framework is open-source and available on GitHub, allowing developers to build their own browser agents. However, the most powerful 'Runner H' models are proprietary.

Does H require a Chrome extension?

No, H typically runs in a cloud-based virtual browser environment, though it can be integrated into local workflows via API.

Can H handle multi-factor authentication (MFA)?

H can navigate to MFA screens, but it generally requires a human-in-the-loop to provide the code unless the code is accessible via a connected app like Gmail or Slack.

What integrations are supported?

H natively connects with Slack, Notion, Gmail, Google Calendar, and Google Sheets. It also supports Zapier for connecting to thousands of other apps.

Why trust this page?

This evaluation combines product positioning, pricing analysis, traffic and market signals, and public user sentiment into a single decision-support page. Content is generated editorially — not copied from the vendor's website.

Funding & Company

Founded

2023

Stage

Seed

Total Raised

$220M

Latest Round

Seed (May 2024)

Notable Investors

Accel Amazon Samsung UiPath Inc. Eric Schmidt Xavier Niel Bernard Arnault

H, a Paris-based AI startup, secured a massive $220 million seed round in May 2024 to develop 'agentic' AI models for automating complex digital tasks. This significant early-stage investment from a consortium of top-tier VCs and strategic tech giants like Amazon, Samsung, and UiPath signals strong confidence in the founding team's vision and technical expertise, providing substantial runway for product development.

Full funding report high confidence

Market Signals & Traffic

Estimated visits, global rank, geography, traffic sources, monthly visit trends, and organic search keywords (Similarweb)—on a dedicated page built for depth and search.

Estimated visits: 1,750
Global rank: #10,137,587
Snapshot: Apr 2026
Traffic trend: Surging