HANI2¶
HANI2 is a web interface for human–agent negotiation studies. People negotiate in the browser against automated agents built with negmas, optionally helped by an AI support agent, while HANI2 records every offer, message, click and questionnaire answer. It is used to run the human–agent negotiation and hybrid intelligence tracks of the ANAC competition.
HANI2 is a Python package with a FastAPI backend and a React frontend that ships pre-built inside the wheel. One command starts everything on one port.

Features¶
The full list, with how to turn each one on, is on the Features page.
- Negotiation desk. The partner's offer on the table with its value to you. Accept it, send a counter-offer with an optional message, or end the negotiation; accepting and ending ask for confirmation. Gauges show what each offer is worth, with a tick at your no-deal value, and the history reads like a conversation.
- A layout you control. Full, simple and focus views; drag the dividers to resize panes; enlarge any tool into a window or the full screen and put it back; layouts for tablets and phones.
- Tools. Thirteen built-in tools, from your preferences (for any kind of utility function) to a negotiation plot with the Pareto frontier. Write your own in Python with a small component API; no frontend code needed.
- Support agent. An assistant beside the negotiation. It chats, keeps a status board, highlights and annotates the interface, rearranges and resizes the layout, shows its own panels, fills in offers and, with permission, acts. It finds the participant's tools, reads and uses them, and adds new tools while they negotiate. Sixteen capabilities that admins grant and participants can narrow.
- Competitions. Run the Human-Agent Negotiation competition without an assistant (
--no-support-agent) and the Hybrid Intelligence competition with the built-in assistant or a team's (--support-agent team.py:TeamAgent), with the team's tools (--tools), against the opponent agents you choose (--opponents). - Any opponent. Any negmas or Genius negotiator, LLM negotiators from negmas-llm, your own classes, and named groups of them, chosen per server, per experiment or per link.
- Studies and Prolific. Self-contained experiments, questionnaires, results CSV files, an event log with analytics, and a Prolific flow with practice rounds and completion links.
- Admin console. Live sessions, results, analytics, users and every setting in the browser.
- Autopilot and WebMCP. Let the support agent or any negotiator play the human's seat, and let browser AI agents use the page.
- Any language model. Local models through Ollama or any provider supported by litellm.
- Easy to install.
pip install hani2on Python 3.11 to 3.15, on Linux, macOS and Windows.
Quick start¶
Your browser opens http://localhost:5006/app. Log in as user / userpass, or as admin / adminpass for the admin tools. The same port serves:
| URL | What |
|---|---|
/app |
The negotiation interface for logged-in users. |
/play |
The playground: no login. Prolific participants use /play?PROLIFIC_PID=.... |
/admin |
The admin console. |
The assistant and LLM opponents need a language model. The defaults expect Ollama with qwen3:4b-instruct:

Where to go next¶
- Features lists everything HANI2 does.
- Installation and the quick start walk through a first negotiation.
- The interface describes what participants see.
- Studies and results and Prolific explain how to run an experiment.
- Build a support agent and Build a tool are hands-on tutorials.
- Configuration reference lists every setting and environment variable.
HANI2 is free software under the AGPL-3.0-or-later license. The source is at github.com/autoneg/hani2.