Runner H
H Company · Coding · Productivity
Runner H is H Company's AI web automation agent. You describe a task in plain words, and it drives a browser or desktop on its own, adapting when a page changes.
About Runner H
What Is Runner H
Runner H is a computer use agent built by H Company, a Paris-based AI lab. Instead of recording rigid clicks, you tell it what you want done. It reads the screen, finds the right buttons and fields, and carries the task through to the end. The pitch is simple: the agent handles the tedious multi-step work that web testing and process automation usually turn into hours of maintenance.
The problem it targets is real. Anyone who has built browser automation knows selectors break every time a site ships a redesign. Runner H works from a vision language model that reads the page the way a person does, so when the layout shifts, it usually keeps going. H Company reports that the agent scored 67% on the WebVoyager benchmark in November 2024, ahead of Anthropic Computer Use at 52%.
There is a catch worth knowing before you commit. Runner H started as a private beta on the H Platform, and the platform is developer-first. If you want a consumer app you can open and poke at, this is not that. You meet it through an API, the Studio, a Chrome extension, or a command line tool.
Getting Started
- Create an account on the H Platform and generate an API key from the settings page
- Install a client, either the hai-agents Python or TypeScript package, or use the hai command line tool
- Describe your task as a direct instruction, for example "find the cheapest direct flight from Paris to Tokyo this Saturday"
- Run a session and watch it work live, then read the answer once the session reaches a finished state
- For repeat jobs, record a routine in the free HoloTab browser extension and schedule it to run again
Product Information
A quick look at Runner H's pricing, supported platforms, and performance.
Best for
The users, tasks, and scenarios where this tool fits best.
Users
- QA and test engineers
- Web developers building internal automation
- Operations and BPO teams
- Product teams running onboarding flows
Tasks
- Running end-to-end e-commerce checks
- Filling multi-step verification forms
- Testing a web app after a release
- Collecting data from pages
- Automating a recurring browser chore
Scenarios
- A redesign breaks your old scripts
- A long research task spans several sites
- You need the same form filled on fifty pages
- A finance team verifies new customers
- You want automation without writing code
Key features
No-code automation from natural language
You describe what you want in plain words, and Runner H writes the automation pipeline for you. There is no selector syntax to learn and no scripting language to pick up. The agent treats your sentence as the spec, then works out the clicks, typing and navigation that get it there.
Self-repairing automation
Runner H adapts to interface changes on its own and repairs itself when something breaks. This is the part that saves the most time, since brittle scripts that die on every site update are the main reason browser automation gets abandoned. You spend your hours on the workflow instead of on upkeep.
Vision language model that reads the screen
The agent runs on Holo, H Company's family of computer use models, starting with a specialized 3B vision language model that locates elements on screen. Give it a screenshot and an instruction like "click the add to cart button," and it returns the coordinates. H Company says this model is far more accurate than much larger generalist ones while costing a fraction as much to serve.
Open-weight models you can run yourself
The Holo family ships at several sizes, from a 27B dense model down to a 0.8B checkpoint for on-device use. The smaller Holo3.1 builds are open weights under Apache 2.0. If privacy or cost pushes you off the cloud, you can download a model and run it on your own hardware, all the way down to a single GPU.
One agent across browser, desktop and Android
Runner H works through whatever interface the task needs. It clicks and types on a screen, writes and runs its own code, and calls APIs or MCP servers, often in a single run. The same model handles web, desktop and Android, so you do not pick a different one for each platform.
Structured output and page reading
You can constrain the agent's answer to a JSON schema, so prices, fields and table rows come back in the shape your code expects. A companion document tool turns a rendered page into clean Markdown, with headings, tables and equations kept in reading order.
HoloTab browser extension
HoloTab is a free Chrome extension that runs the agent right in your tab with no setup, a handy route to no-code automation. You tell it what to do, or record a routine once and re-run it later. Schedules let a recorded routine fire on its own, which suits daily checks and recurring lookups.
Pros and cons
Pros
- Describe a task in plain language instead of maintaining selectors, which removes the biggest upkeep cost in browser automation
- The agent self-repairs when a page changes, so scripts survive redesigns that would break older tools
- Small, specialized models keep serving costs far below large generalist models, and H Company publishes the benchmark comparisons
- A free tier with language models you can run locally means you can test the stack before spending anything
- One model covers web, desktop and Android, so a single setup reaches more places
Cons
- The platform is built for developers, so a non-technical user will not find a simple point-and-click app to start with
- Pricing rides on token usage, which is harder to budget for than a flat monthly fee when tasks vary in length
- The strongest models sit behind the paid tier, so heavy use on the free tier runs into rate limits
- Automation of live sites depends on those sites staying reachable, and captchas or hard logins can still stall a run
Frequently asked questions
Runner H automates multi-step work in a browser or on a desktop. Common jobs cover end-to-end web testing, form filling and verification, data collection from pages, and repetitive chores across several sites. You describe the task, and the agent carries it out.
Related content
Explore related tools, skills, and articles for Runner H.
Runner H Alternatives
Base Code
Base44 (Wix) · Coding · ProductivityBase Code is a shared cloud development environment from Base44. Teammates describe changes to an existing GitHub codebase in chat, preview them live and send them back as pull requests.
Google EmbeddingGemma 2
Google DeepMind · Writing · Image · CodingGoogle EmbeddingGemma 2 is a 740M-parameter open model that turns text, code, images, audio, and video frames into vectors in one 768-dimension space, running on phones and laptops.
GitGlow
GitGlow · Coding · ProductivityGitGlow is a local-first desktop Git client that lets independent developers review their own code and their AI agents' changes in one place before shipping.