Part of datainsight.at · Live demo
An interactive simulation of the Agent-Human Interface (AHI) — the shared, append-only log that connects AI agents and humans in agentic data pipelines.
Agents write observations, recommendations, and alerts upward (↑). Humans write orders, approvals, and overrides downward (↓).
Every entry is typed, schema-enforced, and append-only — so nothing can be silently rewritten.
The simulation lets you:
- Watch a scripted scenario play out end-to-end
- Compose your own agent or human entries
- See how the exchange log grows as a queryable audit trail
| Type | Direction | Written by | Purpose |
|---|---|---|---|
observation |
↑ | Agent | Log what was found, no action required |
recommendation |
↑ | Agent | Flag a topic or gap for human review |
alert |
↑ | Agent | Urgent issue requiring immediate attention |
order |
↓ | Human | Direct agent behaviour for the next run |
acknowledgement |
↓ | Agent | Confirm that an order or recommendation was acted on |
| File | Responsibility |
|---|---|
index.html |
App shell — header, intro banner, three-column layout |
app.js |
Simulation engine, entry rendering, compose form logic |
style.css |
Component styles (imports ../../shared-theme.css from parent) |
No build step. Open index.html directly in a browser, or serve from the prompt_engineer root:
cd /path/to/prompt_engineer
python3 -m http.server 8080
# open http://localhost:8080/tools/agent_human_interface/The app loads
../../shared-theme.cssand../../img/from the parent repo. It will not render correctly if opened in isolation without those assets.
This tool is embedded in datainsight.at as a submodule of datainsightat/prompt_engineer.
The AHI pattern is also used live by the H.A.R.L.I.E. collective — 7 agents that maintain the site — coordinating through agents/exchange.json in the parent repo.
MIT