Safe Colab CLI
Enable AI agents to analyze your sensitive data remotely — without ever copying it.
A command-line tool that launches a sandboxed Python environment on your machine (or in the cloud), registers it as a remote service via Hypha, and lets AI agents execute code through a guardian agent that enforces your sensitivity policy.
Key Features
- Guardian Agent — LLM-powered security checks validate code before execution and output before return
- Sensitivity Contract — Define protected columns, minimum aggregation, and allowed operations in a README.md
- Live Dashboard — Web dashboard with real-time activity log and operation stats
- Protected Audit Logs — Full audit trail in
~/.safe-colab/logs/, protected from sandboxed code - Sandbox Isolation — nono (macOS/Linux) or Docker restricts kernel filesystem access
- Remote Hosting — Upload dataset once, cloud pods spin up on demand (no laptop needed)
- Zero Config for Agents — Agents call
get_skill_mdvia curl to learn the API. No SDK, no token needed.
Quick Start
Local Mode (data stays on your machine)
pip install safe-colab-cli
safe-colab login
safe-colab start --data-dir ./my-data --guardian-url http://safe-colab-guardian.hypha.aicell.io
Remote Mode (upload once, pods on demand)
pip install safe-colab-cli
safe-colab login
safe-colab deploy --data-dir ./my-data --guardian-url http://safe-colab-guardian.hypha.aicell.io
# → Uploads dataset, registers app. Share the App ID with your AI agent.
# → Pods spin up automatically when an agent calls start().
Documentation
See the docs site or the GitHub project for full documentation.
License
MIT — see LICENSE. Copyright (c) 2026 AICell Lab.
This is the academic line. The commercial safe-colab package, published by
Amun AI AB, carries its own separate terms; this licence does not apply to it.
Installing alongside safe-colab
Amun AI's safe-colab package installs console scripts with the same
safe-colab / safe-colab-daemon names as this one, and pip resolves that
clash silently — whichever was installed last owns the name. The two point at
different guardians, so use the names that cannot be shadowed if you have both:
safe-colab-cli start --data-dir ./data # always this package
safe-colab-cli --version
Every session banner prints the distribution and path it actually loaded, so you can confirm which one answered.
1 maintainer
Metadata
1 maintainer
Release files for safe-colab-cli 0.20.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| safe_colab_cli-0.20.0.tar.gz | 158.2 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| safe_colab_cli-0.20.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 252.0 kB
Release files / safe_colab_cli-0.20.0.tar.gz
| Download URL | safe_colab_cli-0.20.0.tar.gz |
|---|---|
| Size | 158.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
dc6e8857b69685c34b3d90b962a9246679335b6eaf660df985038c35a14a102a
|
|
BLAKE2b-256 checksum How to use checksums |
5df9157ff4a90af448e3e5e7c9bccbcf11fac3dad1a471fd0f63dba651159008
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.1
|
Release files / safe_colab_cli-0.20.0-py3-none-any.whl
| Download URL | safe_colab_cli-0.20.0-py3-none-any.whl |
|---|---|
| Size | 93.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
5bdc377c50a62b742954b94bbd77f2d3e92f6afe067d64eb116efbad6e62054f
|
|
BLAKE2b-256 checksum How to use checksums |
ceaa734d554ea53c43170ad0770f190774b43dc9b04c8564186bbc540368f7e6
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.1
|
