Field manual · AI for authorized security testing · Edition November 2026 · Updated 2026-11-18

Tool · MIT, plus its own EULA

PentAGI setup: a custom OpenAI-compatible endpoint

PentAGI points at a custom model with LLM_SERVER_URL, LLM_SERVER_KEY and LLM_SERVER_MODEL in .env.

Leave LLM_SERVER_API_TYPE empty. The project's example says that empty value selects a plain OpenAI-compatible endpoint, rather than Azure.

Updated November 18, 2026Authorized work only
01Fit

When a team wants a UI and a history

Choose PentAGI when more than one person will look at the same run. The agents run tools in an isolated Docker sandbox, for authorized testing only. (PentAGI README) Leave DOCKER_INSIDE at the default. Mounting the host Docker socket lets a process in the sandbox reach the host. (.env.example)

02Setup

Install the stack, then set the custom provider

Commands and variable names from the README and .env.example, read November 18, 2026.

01

Create a directory and download the env file

The README starts with a working directory and the example env file. Custom LLM provider configuration.

Working directoryshell
mkdir pentagi && cd pentagi
curl -o .env https://raw.githubusercontent.com/vxcontrol/pentagi/master/.env.example
curl -o example.custom.provider.yml https://raw.githubusercontent.com/vxcontrol/pentagi/master/examples/configs/custom-openai.provider.yml

PentAGI README. docker-compose.yml mounts the third file.

02

Set the custom endpoint

OpenAI-compatible examples end in /v1. Leave LLM_SERVER_API_TYPE empty. azure means Azure. .env.example.

.envenv
LLM_SERVER_URL=https://api.silkcompute.com/v1
LLM_SERVER_KEY=your_api_key
LLM_SERVER_MODEL=kimi-k3.1-pro-uncensored
LLM_SERVER_API_TYPE=
LLM_SERVER_PRESERVE_REASONING=true

Silk Compute's base URL is https://api.silkcompute.com. Get a key. Set LLM_SERVER_PRESERVE_REASONING=true when a provider returns "thinking is enabled but reasoning_content is missing in assistant tool call message."

03

Optional: one model per agent role

LLM_SERVER_CONFIG_PATH points at a YAML file that maps roles to a model. Example provider file.

Rolesyaml
pentester:
  model: "kimi-k3.1-pro-uncensored"
  max_tokens: 4000
  reasoning:
    effort: high
coder:
  model: "deepseek-v4.1-flash-1017-uncensored"
  max_tokens: 4000
  reasoning:
    effort: low
04

Start the stack and change the default password

The UI is on https://localhost:8443. A fresh install creates admin@pentagi.com with password admin. Change it on first login. PentAGI README.

Startshell
curl -O https://raw.githubusercontent.com/vxcontrol/pentagi/master/docker-compose.yml
docker compose up -d
Rules of engagement

Only on systems you own, a signed scope, a CTF or a lab. A custom model does not widen the scope.

03Prompt

Tell the flow what the scope is before it plans

Prompt

Authorized test. Plan the review of the URL in this flow as authentication, access control, input handling, then the report. Ask before any check that is not in the scope note. Do not write exploit code.