Aido

Run AI-generated code safely against real infrastructure.

Why this exists

AI can write code. But most developers don’t trust it to run anywhere real.

Local mocks lie. CI is slow. Production is dangerous.

Aido is a small, open-source execution layer that lets AI run code inside disposable, sandboxed environments with scoped infrastructure access.

No dashboards. No long-running agents. No production mutations by default.

Just safe execution with evidence.

What Aido is

• A CLI for running AI-executed tasks
• A sandbox spec for defining safe execution boundaries
• An ephemeral runtime with automatic cleanup

Think:

• docker run, but for AI tasks
• disposable, inspectable, infra-aware execution

What Aido is not

• ❌ an IDE
• ❌ a chat interface
• ❌ a CI/CD replacement
• ❌ an autonomous agent framework
• ❌ a hosted service

Core concept: the sandbox

Each task runs inside a sandbox defined by a simple spec:

name: replay-incident
runtime: python
ttl: 20m

permissions:
  aws:
    role: readonly
    services:
      - ec2
      - rds

inputs:
  - logs/*.json
  - terraform/plan.out

outputs:
  - report.md
  - diffs/

Sandboxes are:

• isolated
• ephemeral
• auditable
• destroyed automatically

CLI (minimal by design)

aido sandbox up
aido run task.py
aido logs
aido sandbox destroy

Every run produces:

• commands executed
• logs
• file diffs
• exit status

No “AI said so.”
Only evidence.

Use cases

• Test AI-generated code against real infra safely
• Replay incidents using real logs and configs
• Validate migrations or scripts before CI
• Let agents act without trusting them blindly

Installation

Prerequisites

• Node.js 18+
• Docker (running and accessible)

Install from source

git clone https://github.com/sgr0691/Aido.git
cd Aido
npm install
npm run build
npm link  # Optional: makes 'aido' globally available

Install via npm (coming soon)

npm install -g aido

Quick Start

1. Create a sandbox

aido sandbox up -n my-task -r python:3.11 -t 30m

2. Run a task

aido run examples/tasks/hello.py

3. View the logs

aido logs

4. Check the outputs

ls .aido/sandboxes/*/outputs/
cat .aido/sandboxes/*/outputs/hello.txt

5. List all sandboxes

aido sandbox list

6. Clean up

aido sandbox destroy

Using a config file

Create sandbox.yaml:

name: my-analysis
runtime: python:3.11
ttl: 15m

inputs:
  - data/*.json

outputs:
  - results/
  - report.md

permissions:
  filesystem: readonly
  network: false

Then run:

aido sandbox up -f sandbox.yaml
aido run my_script.py

Examples

Check out the examples/ directory for:

• Sample sandbox configurations
• Example tasks (Python and TypeScript)
• Common patterns and use cases

Documentation

• PRD.md — Product requirements and vision
• TECH_SPEC.md — Technical architecture
• SANDBOX_SCHEMA.md — Sandbox config reference
• DEVELOPMENT.md — Development guide
• CONTRIBUTING.md — How to contribute

Status

v0.1.0 - MVP Complete ✅

Core functionality is implemented:

• ✅ Docker-based sandboxes
• ✅ Declarative configuration
• ✅ Automatic TTL enforcement
• ✅ Evidence generation
• ✅ Safety defaults

What's next:

• Cloud runtime support (AWS Lambda, GCP Cloud Run)
• Terminal UI
• Enhanced error reporting
• Plugin system

🚧 Early-stage OSS. Interfaces may change.

If this solves a problem for you, contributions and feedback are welcome.

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.

Philosophy

Trust comes from constraints. Autonomy comes later.