hai-qa-via-cli

Use cases and integrations for H Company's Computer-Use Agents, built on the hai-agents SDK.

Documentation  ·  Get an API key  ·  Python SDK  ·  TypeScript SDK  ·  H Company

Build with H's Computer-Use Agents, powered by our harness and VLM. A Computer-Use Agent sees the screen and decides what to click, type, and scroll, just like a person would. You describe a task in plain language; H provisions the environment, runs the agent, and returns the result. It's the way in when the work lives behind a UI with no API to call.

The hai-agents SDK gives you programmatic access to our agents from a few lines of Python or TypeScript.

This repo is a tour of three ways to build with the SDK, from a one-line prompt to production integrations:

1. Discover. Describe a task in plain language and watch the agent run it.
2. Build with the skill Let the /hai-agents skill scaffold reusable SDK code for you.
3. Integrate Call the agent as an MCP tool from your coding workflow (Claude Code, Hermes Agent, Codex, NemoClaw).

Quickstart

git clone https://github.com/hcompai/computer-use-agents-demos.git
cd computer-use-agents-demos
uv sync
cp .env.example .env   # add your HAI_API_KEY from https://platform.hcompany.ai/settings/api-keys

1. Discover

Use the SDK directly. The example below is a single Python file: a few lines of the hai-agents SDK wrapped around a plain-language prompt. You write the prompt, the SDK runs the agent, and you've seen a CU agent in action.

Just describe the task in plain language and let the agent run it:

"Searches for "Random Access Memories" by Daft Punk on Amazon, adds it to the shopping cart."

https://github.com/user-attachments/assets/aa7473f8-9666-4640-ac34-6255ab67aa6d

→ Runnable code: examples/add_to_cart/add_to_cart.py (Python)

2. Build with the skill

The /hai-agents skill plugs into your coding agent and carries everything it needs to write SDK code for you: the full API knowledge, the auth and connection patterns, and enough context to write working SDK code quickly, in Python or TypeScript. You get a packaged expert writing your custom code, and an artifact you can deploy anywhere Python or Node runs, independent of the coding agent that scaffolded it. Iterate on prompts to grow your CU agent library.

This repo also doubles as a Claude Code plugin marketplace, so you can install the skill into your own Claude Code without cloning anything:

/plugin marketplace add hcompai/computer-use-agents-demos
/plugin install hai-agents@hai-skills

Ask the skill to write the SDK code, and it scaffolds a ready-to-run script:

"/hai-agents:hai-agents Generate TypeScript code that navigates to jacquemus.com, searches for the France Jacquemus × Nike football jersey, checks its availability in sizes S and XXL, and reports the results clearly."

https://github.com/user-attachments/assets/d7d82573-22bb-4e2a-a261-bb603bc576f3

→ Generated code: examples/product_availability/src/index.ts (TypeScript)

For deeper SDK patterns beyond what the skill scaffolds (custom tools, max_steps / max_time_s budgets, exhaustive sweeps), see examples/counterfeit_detection.

3. Integrate

Now put the agent where you already work. MCP turns it into a first-class tool inside your coding workflow, Claude Code below, plus Hermes, Codex, NemoClaw, or anything that speaks MCP. One agent, many hosts, wiring is the only difference.

Example with Claude Code: