Evo OS is a robust, cost-aware, and self-healing autonomous AI development framework designed to run locally with Docker sandboxing.
What is Evo OS?
Evo OS is an "Agent-First" development environment. Unlike simple coding assistants, Evo acts as a full engineering team. It plans the architecture, writes the code, verifies it using static analysis (AST), executes it in a sandbox to check for runtime errors, and fixes its own mistakes—all autonomously.
It addresses the critical flaws of current AI agents: Infinite Loops, Hidden Costs, and Environment Pollution.
Key Features
BudgetGuard™ (Real-time Cost Control)
Evo tracks token usage in real-time and converts it to actual currency (JPY/USD). It strictly enforces budget limits per run (e.g., stops immediately if it exceeds 50 JPY), preventing unexpected API bills.
Smart Healing & Loop Detection
Uses a multi-stage healing process (Patching -> Rewriting). Crucially, it includes logic to detect healing loops. If Evo gets stuck trying to fix the same error twice, it intelligently adapts its strategy or moves forward to prevent stalled processes.
Docker Sandboxing
All code execution happens inside a secure evo-sandbox Docker container. This ensures:
Security: No risk of malicious code running on your host (e.g., rm -rf / protection).
Consistency: Pre-installed heavy libraries (numpy, pandas, qiskit, scikit-learn, etc.) ensure fast and reproducible execution.
Kit System (Self-Expansion)
Evo selects specialized "Kits" (YAML-based domain knowledge) based on your prompt (e.g., "Make a Chrome Extension", "Create a Mahjong Game"). It can even generate new Kits for itself to learn new technologies on the fly.
Architecture
The system operates on a microservices-like architecture orchestrated by the agent_core:
Planner (Architect Service): Breaks down user prompts into logical implementation phases.
Coding (Workspace Manager): Generates code and manages Git commits for every phase.
Verification (Verifier Service): Performs AST-based static analysis and security checks.
Healing (Healer Service): Autonomously fixes bugs using verify/runtime logs.
Runtime Test (Docker): Executes code in the sandbox to ensure it works.
Self-Improvement (Data Recorder): Logs successful tasks for future fine-tuning.
The project does not claim that AI has consciousness or emotions.
What it provides is a behavioral simulation layer that lets a large language model respond with consistent persona states, memory, and long-term traits.
It’s the same difference as:
• a physics engine is not real physics,
• but it produces behavior consistent enough to be useful.
Users aren’t asking the model to “have feelings.”
They want predictable, stateful, evolving behavior for interactive applications.
Whether AI has consciousness is a philosophical question.
What I am building is an engineering solution.
I built this blackjack game using AI in about 3 hours (design + debugging).
Tech stack: Flask, SocketIO, Redis, simple rule-based AI player.
Feedback is welcome!
Evo OS is a robust, cost-aware, and self-healing autonomous AI development framework designed to run locally with Docker sandboxing.
What is Evo OS?
Evo OS is an "Agent-First" development environment. Unlike simple coding assistants, Evo acts as a full engineering team. It plans the architecture, writes the code, verifies it using static analysis (AST), executes it in a sandbox to check for runtime errors, and fixes its own mistakes—all autonomously.
It addresses the critical flaws of current AI agents: Infinite Loops, Hidden Costs, and Environment Pollution.
Key Features
BudgetGuard™ (Real-time Cost Control)
Evo tracks token usage in real-time and converts it to actual currency (JPY/USD). It strictly enforces budget limits per run (e.g., stops immediately if it exceeds 50 JPY), preventing unexpected API bills.
Smart Healing & Loop Detection
Uses a multi-stage healing process (Patching -> Rewriting). Crucially, it includes logic to detect healing loops. If Evo gets stuck trying to fix the same error twice, it intelligently adapts its strategy or moves forward to prevent stalled processes.
Docker Sandboxing
All code execution happens inside a secure evo-sandbox Docker container. This ensures:
Security: No risk of malicious code running on your host (e.g., rm -rf / protection).
Consistency: Pre-installed heavy libraries (numpy, pandas, qiskit, scikit-learn, etc.) ensure fast and reproducible execution.
Kit System (Self-Expansion)
Evo selects specialized "Kits" (YAML-based domain knowledge) based on your prompt (e.g., "Make a Chrome Extension", "Create a Mahjong Game"). It can even generate new Kits for itself to learn new technologies on the fly.
Architecture
The system operates on a microservices-like architecture orchestrated by the agent_core:
Planner (Architect Service): Breaks down user prompts into logical implementation phases.
Coding (Workspace Manager): Generates code and manages Git commits for every phase.
Verification (Verifier Service): Performs AST-based static analysis and security checks.
Healing (Healer Service): Autonomously fixes bugs using verify/runtime logs.
Runtime Test (Docker): Executes code in the sandbox to ensure it works.
Self-Improvement (Data Recorder): Logs successful tasks for future fine-tuning.