situAgent

Nothing Leaves the Room !

[ local · isolated · private ]

An open-source AI coding agent that runs entirely on your hardware, inside an
isolated container, with --network=none enforced by the OS kernel.
Your source code physically cannot leave the machine — not a promise, a constraint.

View on GitHub See how it works
situ runtime
$ situ
  LM server : POD  (llama.cpp official image)
  Mounted   : ./my-project
  Mode      : RESTRICTED
 
> Refactor the authentication module to use JWTs.
  Analyzing auth.py... writing changes...
MIT License · Apple Silicon NVIDIA CUDA AMD ROCm
Security

No Exfiltration. By Design.

Most AI tools promise "we don't train on your data." SITU makes exfiltration technically impossible. By default, the container runs with --network=none — no network interface exists at all. The agent cannot call home, contact an API, or send a single byte to the outside world, regardless of what the model or user instructs. Not a policy. A namespace constraint the kernel itself enforces.

"Not a privacy policy. A kernel constraint."

situ security testing
% situ --test
[PASSED] LM container is reachable  http://127.0.0.1:8080/
[INFO]   Model in use               gemma-4.gguf
[PASSED] External HTTP is blocked   http://example.com
[PASSED] External HTTPS is blocked  https://example.com
[PASSED] External DNS is blocked    example.com
[PASSED] External TCP is blocked    8.8.8.8:53
Privacy

Leave No Trace

Every SITU session run is an isolated, short-lived container. When you exit, the pod and all its containers are automatically destroyed — no logs retained, no conversation history stored, no residual state. Your context, prompts, and partial outputs vanish completely. What happens in SITU stays in SITU. Full stop.

"The session ends. So does everything in it."

situ session lifecycle
situ> /quit
Shutting down session...
Stopping containers... done
Removing pod... done
 
✓ No logs retained
✓ No conversation history
✓ Residual state: zero
Isolation

Strict Workspace Isolation

SITU only sees what you define. Nothing else exists inside the container. Your home directory, SSH keys, credentials, environment variables, and the rest of your filesystem are completely invisible to the agent. You define the blast radius before the session even starts — and the container boundary enforces it.

"It sees exactly what you show it. Nothing more."

filesystem scope
$ ls ~/
projects/ .ssh/ .env Documents/
 
$ cd ./my-project && situ
✓ Mounted: ./my-project
 
Agent-visible filesystem:
✓ ./my-project
✗ ~/.ssh (not visible)
✗ ~/.env (not visible)
✗ ~/Documents (not visible)
Performance

Your Hardware. Your Rules.

SITU runs on any machine that can run an AI model: a developer MacBook on CPU, an AMD workstation using ROCm, or an NVIDIA GPU rig with CUDA. No cloud account required. No token quota. No rental fees. SITU starts its own secure llama container, but you can also point it to your own llama.cpp instance or LM Studio — the choice is yours. Better hardware means larger models and faster inference, and the ceiling is yours to raise.

"Your hardware. Your model. Your rules."

hardware detection
$ situ --detect
 
✓ Apple M4 Pro detected
✓ Backend: llama.cpp (Metal)
✓ RAM: 48 GB
✓ Model: Qwen2.5-Coder-32B
 
Tip: Use CUDA backend on NVIDIA
Use ROCm backend on AMD
All backends run locally.

Why developers choose SITU

Network Off by Default

The agent has no internet access unless you deliberately switch to NETWORK mode. Sensitive work stays isolated. When you need to fetch documentation, you make a conscious, explicit choice.

100% Open Source (MIT) — Audit It Yourself

Every container definition, every shell script, every line — published under the MIT License. Real transparency means the code speaks for itself. Read it, verify it, run it - securely.

Free. No Subscription. No Surprises.

No per-seat license, no monthly billing, no token quota. The only cost is the hardware you already own — run unlimited sessions on unlimited projects, forever.

Up and Running in 15 Minutes

Three simple steps. The entire stack stays on your machine.

1

Clone & Build

$ git clone \
  github.com/ndburn/SITU-Agent
$ ./build.sh
 
✓ Container built
2

Download a Model

$ curl \
  gemma-4-E4B-it-Q4_K_M.gguf
 
 
✓ Model ready
3

Run the local agent

$ situ \
  -q "Create an API endpoint
  for user authentication"
 
✓ Ready for takeoff.
Full installation walkthrough Knowledge Base · First Steps

Type situ Anywhere. Start Coding.

SITU follows you wherever you work. No configuration needed per project. No plugin to install in your editor. Just a terminal and your project directory.

$ cd ~/fintech-core
$ situ
✓ RESTRICTED mode — network severed
✓ Mounted: ./fintech-core
 
situ> Add rate limiting to the /transfer endpoint.
Reading transfer.py... analyzing...
Writing rate_limiter.py...
Updating transfer.py...
✓ Done. 2 files changed.
 
situ> Write unit tests for the new limiter.
Writing test_rate_limiter.py...
✓ 8 tests written. All pass locally.
 
situ> /quit
Session ended. Pod destroyed. No trace.

Keep your code in the room.

Free, open-source, MIT-licensed. Up and running in 15 minutes — no cloud account, no subscription, no surprises.

View on GitHub