H Platform
Also known as: Runner H, H Company, HoloTab
H Company's computer use agents: the H Platform Agents API runs agents in the browser and on desktops on its Holo models, with HoloTab and HoloDesktop.
H Company, a Paris AI lab, builds computer use agents on its own Holo vision language models. Runner H, its first flagship agent, is no longer presented as a product. Its page now redirects to a newsroom post, and the company sells H Platform instead. The Agents API runs managed agents that see the screen, locate interface elements and act in browsers and on desktops, in H's hosted environments or in the customer's own browser or desktop, with the Holo model of the customer's choice from H's family.
Developers define agents with environments, skills, instructions and subagents, give them custom tools and vault credentials, and run them in sessions that can be watched, steered, paused and resumed, scheduled on a cron, or started from the API, the hai-agents Python and TypeScript SDKs, a CLI, MCP clients and n8n. Every session keeps a complete event record and replay. The API runs from EU or US endpoints, and the enterprise offering adds on premise and sovereign cloud deployment, regional data residency, zero retention, SOC 2 Type II and ISO 27001. HoloTab, a Chrome extension, HoloDesktop, an on device CLI, and open Holo weights are free to try.
H offers no roles or SSO, no memory kept across sessions and no way to test agents against scored cases, and the agent and skill catalog shows no named entries.
Vendor details
Canonical URL
https://hcompany.ai
Category
Browser / computer-use agent
Company status
independent
Use cases & customers
In practice
A finance team faces slow, error prone onboarding. An H Platform agent pre fills multi step verification forms across portals, handles document uploads, and completes the workflow by seeing and clicking each interface element, freeing analysts for higher value review.
A recruiter needs candidates sourced and logged. From a single natural language prompt, an H Platform agent gathers resumes from the web, navigates the applicant tracking system, and updates the customer relationship management records without brittle scripts to maintain.
A quality assurance team must test an e-commerce funnel across payment methods. An H Platform agent runs each scenario as a real user would, seeing the screen and acting in the browser, and the session replay shows where a run failed.
Sources & related URLs
Agentic Index coverage score
9.0 / 14 capabilities · 64%
| Integrations & Tool Calling | Full |
|---|---|
|
Agents can be given custom tools whose results the customer's code returns to the session, beside operating any interface directly. Agents also run from n8n and MCP clients. Holo agents can call MCP tools and the customer's own APIs, and vaults let an agent sign in to the services it works on without passing secrets. SourceH Company, hub.hcompany.ai custom tools, vaults and Models API docsread 2026-10-05 |
|
| Workflow Orchestration | Full |
|
Agents are defined with environments, skills, instructions and subagents, parallelize work across subagents, and run in sessions that can be paused, resumed and scheduled. The enterprise platform orchestrates workflows across systems and switches to computer use for steps without an API. One manager agent can split a task across specialist subagents, and a single Holo agent can use both a browser and a Python runner. The final answer can come back as typed JSON checked against a schema. SourceH Company, hub.hcompany.ai Agents API and Models API docsread 2026-10-05 |
|
| Knowledge Grounding & RAG | Not documented |
|
Agents read interfaces and documents at run time, and H keeps no store of the customer's knowledge for them to search. The Models API offers document OCR as a model capability, turning a rendered page into Markdown, and skills hold reusable instructions. SourceH Company, hub.hcompany.ai llms.txt and skills docsread 2026-10-05 |
|
| Human Oversight & Guardrails | Partial |
|
A running session can be watched, paused, resumed, steered with messages or forced to answer, and two factor prompts are handed to a person. On the enterprise platform, agents stay within authorized boundaries and escalate to a person when something falls outside scope. An approval step before an action goes through appears only in an enterprise case description, not as a setting the customer can turn on. SourceH Company, hub.hcompany.ai observe and steer and hcompany.ai/enterpriseread 2026-10-05 |
|
| Security, Identity & Governance | Partial |
|
H reports SOC 2 Type II and ISO 27001, with a Vanta trust center at trust.hcompany.ai, plus vaults for credentials and redaction before data leaves the perimeter. There are no roles, SSO or other access model. Multi tenant deployments use dedicated encryption keys and API keys with strict logical isolation. Data never passes through third party models, since H runs the whole inference chain. SourceH Company, hcompany.ai/enterpriseread 2026-10-05 |
|
| Observability & Auditability | Full |
|
Each session keeps a complete paginated event record of what the agent observed and did, with full logs and session replay of every action, and enterprise exports are audit ready. A status endpoint reports state, step count and token use, a session can be shared through a public Agent View link, and screenshots and files from a session can be downloaded. On the enterprise platform, every action is timestamped and the reasoning behind each decision is kept. SourceH Company, hub.hcompany.ai observe and steer and sessions docs, and hcompany.ai/enterpriseread 2026-10-05 |
|
| Memory & State Persistence | Not documented |
|
Sessions carry their own history, and a customer can start a cloud browser from a saved profile of cookies and storage. That is browser state, and there is no memory an agent reads from one session to the next. A paused session keeps its state and resumes where the agent stopped. SourceH Company, hub.hcompany.ai sessions and browser profile docsread 2026-10-05 |
|
| Deployment & Data Residency | Full |
|
The Agents API runs from an EU or a US base URL, agents can run in the customer's own browser or desktop, Holo models can be self hosted, and the enterprise offering lists regional data residency, zero retention, on premise and sovereign cloud. On premises and private cloud are assessed case by case, and the default multi tenant setup keeps each customer logically isolated. Open weight Holo4 and Holo3.1 models can be served locally with llama.cpp or vLLM, and the browser can route through an H managed proxy or the customer's own. SourceH Company, hcompany.ai/enterprise and hub.hcompany.ai docsread 2026-10-05 |
|
| Prebuilt Agents, Templates & Packs | Partial |
|
A catalog gallery of agents and skills from H loads in the browser, but no entry or job is named. The agents-api-demos repository offers examples. Skills are reusable instruction fragments an agent draws on when a task matches. H describes KYC onboarding, order to cash and procurement deployments, with results such as a 50 to 70% cut in time to compliance. SourceH Company, hub.hcompany.ai skills and hcompany.ai/enterpriseread 2026-10-05 |
|
| Triggers & Channel Coverage | Full |
|
Schedules create sessions on a cron expression and can be paused and resumed, beside sessions started from the API, SDKs, CLI, MCP clients, n8n and the platform, with webhooks for notifications. Webhooks are signed, and their secret can be rotated without a failed delivery. Agents can also be started from Claude, Cursor or any MCP host. SourceH Company, hub.hcompany.ai schedules, webhooks and MCP docsread 2026-10-05 |
|
| Model Flexibility & Routing | Not documented |
|
An agent's model is any Holo model id from H's own family (default holo3-122b-a10b), and there is no choice of provider. Holo4 and Holo3.1 have open weights, and H runs the whole inference chain itself. SourceH Company, hub.hcompany.ai agent create and Models API docs, and hcompany.ai/enterpriseread 2026-10-05 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
There is a REST API (agents, environments, sessions, skills, vaults, schedules), hai-agents Python and TypeScript SDKs and a CLI, and agents run over MCP. The Models API serves Holo models with open weights. It is OpenAI compatible at api.hcompany.ai/v1, with function calling and structured outputs. A skill also teaches coding assistants the H APIs. SourceH Company, hub.hcompany.ai API reference and Models API docsread 2026-10-05 |
|
| Testing, Debugging & Optimization | Partial |
|
Sessions and single events take feedback through the API to track agent quality, and replays show where a run failed. There are no scored test cases or checks an agent must pass, and the QA testing use case tests the customer's application, not the agent. SourceH Company, hub.hcompany.ai session and event feedbackread 2026-10-05 |
|
| Browser & Computer Use | Full |
|
H Platform agents see the screen, locate interface elements and act in the browser and on desktops, in H's hosted environments or in the customer's own browser or desktop, on the Holo computer use models. On the customer's own machine they drive Chrome or the mouse, keyboard and screen, and they can request one time codes and confirmation links during a login. SourceH Company, hcompany.ai/agents-api and hub.hcompany.ai docsread 2026-10-05 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Agents API 1.5.1 adds shell mode for desktop environments, so an agent can drive a machine through commands: running them, editing files and writing code. A command_approval setting holds each command for human approval unless it matches an allowed_commands list, and on macOS commands run in a sandbox by default that limits writes to the session folder and blocks access to stored credentials. The holo CLI prompts at the terminal to approve, always allow or deny held calls.
Bears on: Human approval / guardrails
View sourceRelease 1.4.0 ships the holo command line tool, which runs H's browser and desktop agents from a terminal in interactive sessions where users send follow ups, pause the agent and correct it. A built in MCP server lets coding agents list H agents and start runs, and one install command adds it to 13 coding agent hosts or connects them to the hosted MCP server instead.
Bears on: MCP / tool calling / API
View sourceRelease 1.2.1 of H's Agents API adds local mode in the Python SDK: Client.local() runs the agent on the customer's own machine with the same sessions API as the cloud, using H's hosted models or the customer's own model server. Later releases in the window added Linux support and the option to serve H's Holo4 models from a customer server.
Bears on: Browser/computer use
View sourcePricing
Free tier; Agents API Builder from $20 a month
What is public
hcompany.ai/pricing publishes Agents API session plans (Free $0, Builder $20 a month, Startup $200 a month, Enterprise custom with negotiated markup) with the first $20 or $200 of usage at model cost, per token Models API rates, and free HoloTab, HoloDesktop CLI and open Holo weights. The earlier Runner H wave rollout is withdrawn: Runner H's page now redirects to a newsroom post.
Free / trial
Free Agents API plan; HoloTab extension, HoloDesktop CLI and open weights free
Lowest paid plan
Builder, $20 a month
Key ambiguities
Sessions per plan are approximate and vary with task complexity and the Holo model used.
Related vendors
- AgentQL — A self healing query language for the web, with natural language…
- AGI, Inc. — AGI, Inc
- Airtop — Cloud browsers for AI agents that compile plain English workflows…
- Asteroid — Healthcare portal integration platform whose supervised browser and…
- Automat AI — AI agents that operate computers visually, sold as Automat Workforce…
- Autotab — General AI agent that learns a workflow from a demonstration and…
Alternatives to H Platform
The closest documented capability profiles to H Platform among browser and computer-use agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Airtop9.0 / 14Adds documented Model Flexibility & Routing
- Autotab8.0 / 14Adds documented Memory & State Persistence
- Axiom.ai10.0 / 14Adds documented Model Flexibility & Routing
- Automat AI11.5 / 14Adds documented Knowledge Grounding & RAG and Memory & State Persistence, among others
- Browser Use11.5 / 14Adds documented Memory & State Persistence and Model Flexibility & Routing
- Narada10.5 / 14Adds documented Knowledge Grounding & RAG and Memory & State Persistence
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded