Skip to main content

Understand and debug agent harnesses from the model's perspective.

Project description

HumanHarness

Understand and debug agent harnesses from the model's perspective.

HumanHarness is a quick and easy tool to let you view an agent's harness yourself.

Simply start the app and set your OpenAI-compatible API to http://localhost:8080/v1 or whatever port you've set.

The twist: instead of using an AI model, there is a web UI which you (probably a human) can visit, and from there, any incoming requests are shown. The full text, system prompt, and chat history. Also, any available tools are shown in the UI as buttons, with parameters you can edit. Finally, when you've finished "thinking" about it, you can type a response back.

The goal here is to be able to look at things like OpenClaw, Pi, Hermes, Claude Code, OpenAI Codex, or any other agentic framework and getting a deeper level of understanding how it works + seeing it from the AI's perspective + being able to have a bit of fun. You can also use it to help you debug model performance in your own agentic harnesses! You can see what the LLM is seeing and optimize parts of the harness (e.g. visible toolset) to help maximise model performance.

Installation

Quick installation:

$ pip install humanharness

Or from source:

$ gh repo clone uukelele/HumanHarness
$ cd HumanHarness/humanharness/frontend
$ npm i && npm run build
$ cd ../..
$ pip install -e .

Starting the server

$ humanharness

Starts the web server at port 8080 by default.

To auto-open in browser:

$ humanharness --open

To set another port / host:

$ humanharness --host 0.0.0.0 --port 8081

Use with an agent

Simply set your base URL to the one that appears in the web dashboard.

For example, if using the default settings:

http://localhost:8080/v1

The chat completion URL would be http://localhost:8080/v1/chat/completions

Features

  • /v1/chat/completions
    • Text responses
    • Viewing images
    • JSON-based tool calling
    • XML-based tool calling
    • Other forms of multimodal content
  • `/v1/responses

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

humanharness-0.1.0.tar.gz (59.7 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

humanharness-0.1.0-py3-none-any.whl (62.0 kB view details)

Uploaded Python 3

File details

Details for the file humanharness-0.1.0.tar.gz.

File metadata

  • Download URL: humanharness-0.1.0.tar.gz
  • Upload date:
  • Size: 59.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.10.20

File hashes

Hashes for humanharness-0.1.0.tar.gz
Algorithm Hash digest
SHA256 713891a0aa3523bde428038963ba4b042aa868662cfc0c3d64f70745a06147c3
MD5 dcf1fdc95bc73af1b0a6344885598f7a
BLAKE2b-256 2cfb25f995fc32277b83c2ed5dc0d35b7ae4f1ea142fcdb1394748285e267ef4

See more details on using hashes here.

File details

Details for the file humanharness-0.1.0-py3-none-any.whl.

File metadata

  • Download URL: humanharness-0.1.0-py3-none-any.whl
  • Upload date:
  • Size: 62.0 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.10.20

File hashes

Hashes for humanharness-0.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 9a718d81bc325b434f2ad72114eb7e4867898d4dc5082c0310d607d391fa9859
MD5 bcb33e1ec4495510e8af0b3db2568376
BLAKE2b-256 11df5352b2385e446a41e7731e4a03699069670f7cd414808c40172733c13fc7

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page