H Company

Automates work across web, desktop, and enterprise systems with configurable agents, integrated tools, and controls for monitoring their runs.

Build with Holo models to process multimodal content and perform computer-use tasks through a familiar API, with flexible hosted or local inference.

Holo models help teams build agents that understand software interfaces and carry out tasks across digital environments.

HoloTab takes everyday browsing tasks off users’ hands, with reusable routines and scheduling to reduce repeated manual work.

Run an agent on your device with Models API inference or a self-hosted model.

Automate work across existing enterprise systems while keeping actions traceable, data controlled, and deployment supported from discovery through production scale.

Understand bot fragility and total operating costs, compare agent alternatives, and get a prioritized retirement shortlist before your next license renewal.

Measure how much of a bot exception queue an agent can complete and get results and a launch proposal in two weeks.

Supports long-context document understanding with multimodal encoders and retrievers available in multiple sizes and variants.

H Platform

H Platform is an enterprise platform for configuring, connecting, running, and monitoring agents across web, desktop, and enterprise systems. It combines a proprietary agent execution engine, workflow orchestration for complex long-horizon tasks, virtual and local environments, and inference on H-hosted models. Teams can use the Agents API and Models API, SDKs, MCP, CLI, and connectors, with reusable skills and specialist subagents. Agents can navigate live web apps visually, perform computer actions such as clicking and typing, and adapt to interface changes. The platform supports live session monitoring and action-by-action replay, alongside security controls including secured tunnels to customer machines, system-level permissions, SSO, projects, and role-based access control.

Category: AI Agent Platforms

• Configure And Deploy Agents
  Set up agents for their tasks and deploy them across the systems where work takes place.
• Run Agents Across Environments
  Use agents across web, browser, desktop, and enterprise systems, with hosted or local environments for the surfaces they need.
• Orchestrate Complex Agent Workflows
  Coordinate complex, long-running tasks while keeping agent building, execution, and evaluation within a single platform.
• Monitor And Replay Agent Runs
  Observe sessions as they happen and review agent actions afterward to understand outcomes and investigate failures.
• Control Access Across Deployments
  Apply system-level permissions and manage access with SSO, projects, and role-based access controls. Secured tunnels connect to customer machines.
• Run Agents On H's Hosted Models
  Use an agent execution engine designed for production alongside inference on models H hosts and operates for agent workloads.
• Automate Tasks Through Visual Interfaces
  Build agents that interact with software through its interface, including systems that do not provide APIs.
• Keep Runs Reliable After UI Updates
  Build agents that can adapt when real interfaces change, reducing reliance on fixed interactions as workflows evolve.
• Run Agents On An Integrated Stack
  Use an integrated runtime and infrastructure stack for agent execution instead of assembling separate components.
• Connect Agents To Existing Tools
  Build with the APIs and SDKs or connect agents to supported tools and existing stacks using one API key.
• Browse Web Apps And Extract Data
  Automate live web workflows and return extracted information in the requested format.
• Reuse Skills And Delegate Tasks
  Equip agents with reusable skills and assign specialized work to subagents when tasks require them.
• Reduce Cost Per GUI Run
  The stated 40% lower cost per GUI run can help reduce model costs for comparable tasks.
• Choose Data Residency And Retention Options
  Use regional data residency and zero-retention mode alongside the listed security certifications.

Models API

An OpenAI-compatible API for using Holo models to process text, images, and documents, with computer-use capabilities for digital environments. The API offers models in multiple sizes, including Holo4 models with 256K context, a rate-limited free tier for Holo3.1, and pay-as-you-go access. Selected models can also be served locally with llama.cpp or vLLM using the same API client.

• Build Computer-Use Workflows Across Environments
  Use Holo models for tasks involving digital interfaces and tools, including desktop, web, and mobile environments.
• Integrate Holo Models With Familiar Tools
  Use an OpenAI-compatible interface and familiar client patterns to connect Holo models to applications and agent workflows.
• Process Text, Images, And Documents
  Send text, images, and documents to Holo models through one API for multimodal workloads.
• Choose Models For Different Workloads
  Select from available Holo model sizes and generations to suit different workloads, using published context lengths and model descriptions to guide selection.
• Start Free And Pay For Model Usage
  Try Holo3.1 through rate-limited free access, then use pay-as-you-go model access with per-token pricing.
• Run Supported Models On Your Infrastructure
  Run selected Holo models with llama.cpp or vLLM and direct the same API client to the local server.
• Keep Submitted Data Out Of Persistent Storage
  The API is described as having zero data retention, and its terms state that processed text and visual data is not stored persistently.
• Locate Interface Elements From Screenshots
  Provide a screenshot and target description, then use returned coordinates to guide an agent's interaction with an interface.
• Convert Rendered Pages Into Markdown
  Extract structured page content as Markdown while preserving headings, tables, equations, and reading order.
• Download Open Weights For Supported Models
  Access Holo4 35B-A3B weights under Apache 2.0 and find open checkpoints and local inference guidance for supported models.

Holo models

Holo is H Company's family of vision-language and action models trained to understand software interfaces and act on instructions. The models support computer-use tasks across digital environments, with capabilities varying by release, including visual grounding, action generation, long-horizon control, and tool use. The family includes models in multiple sizes and architectures. Selected releases have open weights, quantized checkpoints, and options for local inference; models are also available through H Company's Models API.

• Ground Instructions In Interface Actions
  Interpret what is visible on screen and translate task instructions into actions for an agent to perform.
• Operate Across Web Desktop And Mobile
  Use models designed for computer-use tasks across different software environments, rather than only one interface category.
• Combine Screens Code And Tools
  For tasks that span interfaces, Holo4 can use screen actions alongside code and MCP or API tools.
• Manage Multi-Step Workflows
  Support agents as they work through tasks involving multiple actions, progress checks, and changing interface states.
• Process Text Images And Documents
  Build applications that use Holo models to process text, images, and documents through the Models API.
• Run Models Locally
  Use selected checkpoints for local inference, including configurations designed for local or single-GPU deployment. Holo3.1 describes fully local execution that keeps activity on the user's network.
• Download Selected Model Checkpoints
  Access selected Holo model weights through Hugging Face and use them according to their individual license terms.
• Choose A Model For Your Workload
  Select among available Holo releases and sizes, comparing specifications, benchmark results, licenses, and access options.
• Integrate With Function-Calling Agent Stacks
  Use Holo3.1's native function-calling protocols or structured JSON outputs when integrating it into supported agent frameworks.
• Compare Published Model Results
  Review published evaluation results to compare model capabilities on computer-use and interface-grounding benchmarks.

HoloTab

HoloTab is a free Chrome extension that works from a browser side panel to navigate websites and handle everyday tasks such as managing email, scheduling meetings, researching trips, and listing products. Users can give typed or spoken instructions, watch its reasoning as it works, and save tasks as reusable routines that can run on a schedule. HoloTab can summarize or extract page content. It requires no signup, and its terms state that it deactivates on more than 150 sensitive domains.

• Automate Everyday Web Tasks
  Hand off routine browsing work, including email management, meeting scheduling, trip research, and marketplace listings.
• Start Browser Automation For Free
  Install HoloTab from the Chrome Web Store and open it in the browser side panel without signing up.
• Save Tasks As Reusable Routines
  Capture the steps for a task once, then reuse the routine instead of entering the same instructions each time.
• Schedule Routines To Run Automatically
  Set a daily, weekly, or monthly schedule for a routine and let it run in the background.
• Follow The Agent's Progress
  See the active tab and the agent's next planned action as it works through a task.
• Summarize Pages And Extract Content
  Turn content from the active tab into a summary or extract relevant information for a task.
• Deactivate On Sensitive Domains
  The extension disables itself on more than 150 stated sensitive domains across banking, health, HR and payroll, and adult content.
• Give Tasks By Voice Or Text
  Choose between typing and speaking instructions to submit a task.
• Delete Screenshots After Processing
  The product's terms and privacy policy describe screenshots as temporary task data that is deleted after analysis or task completion.

HoloDesktop CLI

HoloDesktop CLI is a command-line tool for running an agent on-device. It supports inference through the Models API or a self-hosted model.

• Run Agents On Your Device
  Execute an agent on-device using HoloDesktop CLI.
• Choose Hosted Or Self-Hosted Inference
  Select Models API inference or use a self-hosted model to run the agent.

Enterprise agent deployment

H Company works with enterprise teams to deploy computer-use agents on their existing systems, including legacy software, mainframes, and desktop applications, without requiring APIs, connectors, or infrastructure rebuilds. The engagement moves from workflow discovery and prototyping on the customer's systems to an agreed proposal, production launch with embedded H Company engineers, and scaling across processes, teams, and geographies. Agents operate directly through software interfaces, with controls for authorized access, human review, traceability, and data protection. Multi-tenant deployment is the default; on-premise and private cloud options are assessed case by case for organizations with specific regulatory or data-sovereignty requirements.

• Automate Work Across Existing Systems
  Deploy agents on the software and infrastructure already in use, including systems that lack APIs or connectors, without rebuilding the technology stack.
• Protect Data And Enforce Boundaries
  Keep agents within authorized boundaries, initiate connections from the customer's side, and redact sensitive screenshot fields before data leaves the perimeter. Regional data residency and zero-retention mode are available.
• Deploy With Embedded Engineers
  Work with H Company engineers who learn the customer's systems, build agents on its infrastructure, and support deployment through launch and scaling.
• Validate Workflows Before Production
  Map a workflow, test a working agent on actual systems, and agree scope, timeline, and success metrics before production launch.
• Review And Replay Every Action
  Inspect timestamped action logs, replay agent runs, review decision reasoning, and export records for regulatory compliance.
• Keep People In Control
  Set review thresholds, begin with read-only access, approve write actions step by step, and route out-of-scope or judgment-based work to people.
• Choose A Deployment Model
  Use the default multi-tenant model with customer-specific keys and logical isolation, or assess on-premise and private cloud options for specific regulatory or sovereignty needs.
• Keep Inference Under Your Control
  Use H Company's end-to-end inference chain without routing enterprise data through third-party models, with on-premise and sovereign cloud options available.
• Work With One Accountable Partner
  Coordinate with one team accountable for the model, orchestration, agents, and deployment outcome, using a stack H Company owns end to end.
• Meet Enterprise Security Requirements
  Use listed security and compliance controls including SOC 2 Type II, ISO 27001, secured tunnels, customer-level permissions, SSO, role-based access control, regional data residency, and zero-retention mode.
• Keep Inference Costs Predictable
  Avoid surprise inference bills and exposure to third-party model pricing changes.
• Move From Scope To Production Faster
  Use a structured deployment path and a single team and stack to iterate from workflow scoping toward production without external dependencies.

Bot Estate Audit

A free, two-week assessment of an organization's automation estate. Using an export of bot names, schedules, run history, and failure logs, plus two workshops, it assesses each bot's fragility and costs, compares bot run costs with the cost of an agent handling the same process, and recommends bots to retire first. The week-two readout includes reusable cost assumptions and a shortlist of bots for a potential paid prototype. The audit requires no access to production systems and carries no commitment beyond the audit.

• Identify Bot Breakage And Maintenance Effort
  See the frequency and causes of bot failures alongside the hours your team spends fixing them, creating a fact base for assessing maintenance demands.
• Compare Bot And Agent Run Costs
  Compare each bot's license, infrastructure, and maintenance costs with the modeled cost of an agent running the same process.
• Prioritize Bots For Retirement
  Get a prioritized list of processes where computer-use agents could take over, including a shortlist of bots to consider for a paid prototype.
• Present Reusable Cost Assumptions
  Use the cost model and its assumptions in a readout designed for the automation lead and finance partner.
• Complete An Assessment In Two Weeks
  Share a bot inventory on day one, attend two workshops during week one, and receive the readout and recommendations in week two.
• Assess Bots Without Production Access
  Provide an orchestrator export rather than access to production systems. The audit is read-only by default and uses an outbound-only connection.
• Evaluate Your Estate At No Cost
  Receive the audit without charge or a commitment beyond it. If findings support it, the proposed next step is a paid prototype on two or three bots.

Exception Queue Sprint

A two-week engagement that puts a live computer-use agent on one of your exception queues in a sandbox. The agent works real exceptions across portals, legacy screens, and documents, while H Company measures how many items it completes and which it routes to a person. The Sprint also provides failure reasons and a launch proposal for the selected queue and two more. Existing bots keep running, and human review thresholds are set by you.

• Test An Agent On Real Exceptions
  See how an agent performs on real work from one of your exception queues, using your systems and data in a sandbox.
• Measure Queue Completion
  Get a measured view of the agent's completion rate, items routed to people, and reasons items were not completed.
• Route Judgment Calls To Reviewers
  Keep people involved where their judgment matters by deciding which items the agent can close and which must go to a reviewer.
• Start Safely In Your Sandbox
  Evaluate the agent in a test environment with a user-level account, read-only operation at first, and write actions only after approval.
• Plan A Production Launch
  Use defined scope, queue volumes, and KPIs to evaluate a potential production launch across the selected queue and two additional queues.
• Get Results In Two Weeks
  Follow a defined two-week process from queue selection and KPI agreement through agent testing to results and a proposed next step.
• Keep Existing Bots Running
  Continue using existing bots for clean, structured work while testing an agent on the exceptions those bots currently send to people.

NeoMME

NeoMME is an open model family of multimodal encoders and retrievers for long-context document understanding. It is available in 260M and 800M parameter sizes, with dense and late interaction retriever variants, under the Apache 2.0 license.

• Understand Long Documents
  Use multimodal encoders and retrievers designed to support understanding of documents with long contexts.
• Choose A Model Variant
  Select from 260M or 800M parameter models and dense or late interaction retriever variants.
• Use Apache 2.0 Licensed Models
  Access the model under the Apache 2.0 license.