Local, Secure & Intelligent

Your agent, ready in ~95 lines of Python. Ready for the Agentic?

Prometheus is built on four pillars — Harness, Loop, Memory, Eval/LLM-Ops — with no frameworks hiding the good parts. Your memory is one SQLite file. Open it. Analyse it. It's yours.

Install it Read the code

watch it think

loop ~95 lines

memory state.db

Four pillars

Nothing is hidden behind a framework.

These are the four boxes the whole project is drawn as. Each one is a directory you can open, not a feature on a pricing page.

Harness

Gateways — terminal, dashboard, voice, Telegram, Discord, WhatsApp — and the tools around one loop.

Loop

~95 lines of plain Python: reason, act, repeat, with two ways to stop. Step through it.

Memory

Semantic, episodic and procedural. A gate decides whether to remember; consolidation decides what to keep.

Eval / LLM-Ops

Deterministic tests and LLM-as-judge, side by side, with a release gate and a trace for every turn.

Get it running

Two ways in.

Just want to run it, install from pip. Want to read the code — which is the point of this repo — clone it instead. Every block below is copy-pasteable.

Just run it

01 install

pip install prometheus

02 talk to it in the terminal

# your first run tells you which key to set
prometheus

03 or open the browser cockpit

prometheus dashboard    # → localhost:7777

Or read the code

04 clone, then install into a venv

git clone https://github.com/Engine-User/Prometheus-AI-Agent
cd Prometheus-AI-Agent
uv venv && uv pip install -e .
cp .env.example .env    # pick a provider, paste ONE key

05 run it from the checkout

uv run prometheus
uv run prometheus dashboard

Use the model you already pay for

One key. Ten providers.

Set PROMETHEUS_PROVIDER and paste the key. One dialect in the loop; a ~60-line adapter handles the rest.

Anthropic

OpenAI

Gemini

DeepSeek

MiniMax

Kimi

GLM

OpenRouter

OpenCode Zen

OpenCode Go

Docs, in the order to read them