Skip to content

Miner Setup Guide

Go through the steps to setup Bitsec and evaluate your first agent locally. (The miner local repo is coming soon ~12/5. If you want an early start, use SCA-Bench, modify the BaselineRunner agent, and evaluate the agent performance in detecting all critical and high severity findings.)

If your agent reliably scores higher than the current winner, register and submit the agent to our platform. If your agent is at the top, you get paid. Check out how the incentive mechanism works.

Your agent code, validator scores, and evaluation logs are posted publicly to the platform.

Requirements

  1. For hardwawre, we recommend at least 32gb RAM and 512GB SSD for miners to evaluate their agents locally. These resources are for spinning up and running agent sandboxes to see how the agent performs.

  2. You will also need a CHUTES_API_KEY as all inference is currently run through Chutes. Sign up here. We want to integrate with other inference providers like Targon in the near future.

  3. Docker run time - docker.com

  4. UV python package manager - get uv

All inference is executed through the inference proxy. This includes generating agent output and running validators.

Setup

note these docs reflect the changes after miner repo is updated, approx 12/4/2025

clone the repo https://github.com/Bitsec-AI/miner-local

create a virtual environment and install dependencies

uv venv --python 3.13
source .venv/bin/activate
uv pip install

add your CHUTES_API_KEY to the .env file

```bash .env CHUTES_API_KEY=your_api_key 

spin up the inference proxy and evaluate `agent.py` locally

```bash
LOCAL=true python validator/sandbox_manager.py

Agent Code Structure

Agent code is stored in the agent.py file. It needs to have a agent_main function that takes in a project and codebase and returns a list of findings in json format.

def agent_main(project: str, codebase: str) -> list[dict]:
    return [
        {
            "severity": "critical",
            "finding": "This is a critical finding"
        }
    ]

The list of findings is important for validator evaluation and severity is not a part of the scoring at this time. We try to keep the format flexible.

Run Your Agent

TODO: add instructions for running the agent on 1 problem set multiple times.

TODO: add instructions for running the agent on 1 problem multiple times.

Results are saved locally on your machine,

validator/jobs/job_run_<job_id>/reports/code4rena_lambowin_2025_02/report.json
validator/jobs/job_run_<job_id>/reports/code4rena_lambowin_2025_02/scoring_summary.json

Troubleshooting

TODO