Self-hosted AI agents that call tools, edit code, and remember.

Bring your own model and GPU. A tool-calling loop with safe edits, checkpoints, persistent RAG memory, a streaming API, and human-in-the-loop approvals — running entirely on your box.

Get started GitHub

agent — code_agent

$ql agent "summarize the README"
›search_code  query=def main
 agents/code_agent.py:133   services/main.py:1
›read_file  path=README.md
 # Quantum Labs
✓done  "# Quantum Labs"

Tool-calling agents

A ReAct loop that searches the repo, reads files, writes changes, and runs commands.

Fully self-hosted

Your model, your GPU. Any OpenAI-compatible endpoint — Ollama, vLLM, or your own pod.

Safe edits

Every write is checkpointed. Roll a session back to any point with one command.

Persistent memory

RAG over past sessions; relevant context is auto-injected into new tasks.

Streaming API

A FastAPI backend with SSE task streams — watch the tool-call loop live.

Human-in-the-loop

Approve or deny file writes and shell commands before they ever run.

Run it in three commands

bash

$ git clone https://github.com/takemewa-collab/QuantumLabs
$ cd QuantumLabs
$ ./dev.sh

Then open http://localhost:3000