doit-fm 1.0.0

Telegram-Controlled File Management Engine

🏋️ Doit — Telegram-Controlled Local Automation Engine

Doit is a production-grade open-source automation agent that runs silently on your laptop and executes real OS-level tasks — controlled entirely through Telegram.

No GUI. No web dashboard. No daily terminal usage. Just Telegram.

---

Features

Category	Capabilities
File System	Create, read, write, delete, move, copy, search, compress, extract, organize, backup, clean
System	CPU/RAM/disk monitoring, process management, shutdown/restart/sleep
Network	Download files, ping, HTTP requests, URL uptime monitoring
Scheduler	Natural language + cron scheduling, persistent, crash-recoverable
Security	Single-user auth, encrypted config, injection protection, tool sandboxing
Resilience	Crash recovery, network loss tolerance, power-loss recovery, auto-restart
Plugins	Dynamic discovery, sandboxed execution, documented API
Observability	Status, tasks, logs, health, audit export (JSON/CSV)

---

Quick Start

pip install doit-agent
doit init

The wizard walks you through:

1. Trust grant (one-time)
2. AI provider & model selection
3. Telegram bot setup
4. Auto-start service installation

After setup: just talk to your bot.

---

Usage (via Telegram)

list files in ~/Downloads
download https://example.com/report.pdf to ~/Desktop
backup ~/Documents to ~/Backups every day at 9am
show system health
kill process 12345
organize ~/Downloads
compress ~/Projects/myapp to ~/Backups/myapp.zip
safe mode          → pause all execution
resume             → resume execution
/status            → agent status
/tasks             → recent tasks
/health            → system resources
/logs              → recent log entries
/export json       → download audit log

---

Architecture

doit/
├── cli/           # Bootstrap, onboarding wizard, CLI commands
├── core/          # Config, constants, cross-platform paths
├── ai/            # Multi-provider AI intent engine
├── telegram_bot/  # Telegram interface, auth, injection guard
├── tools/         # Tool registry + all tool implementations
├── task_engine/   # Async queue, workers, retry, resource guard
├── scheduler/     # Natural language + cron scheduler
├── security/      # Encryption, lock file, authorization
├── persistence/   # SQLite database (tasks, logs, config, audit)
├── plugins/       # Plugin system with dynamic discovery
├── services/      # OS service (systemd/LaunchAgent/Windows)
├── supervisor/    # Process supervision, crash recovery
├── logging_system/# Structured logging to file + DB
└── updates/       # Version check, safe update, rollback

---

AI Providers

#	Provider	Best Models
1	NVIDIA NIM	Llama 3.3 70B, Nemotron Ultra 253B
2	Zhipu AI (z.ai)	GLM-4 Flash (free), GLM-4 Air
3	OpenAI	GPT-4o, GPT-4o Mini
4	Anthropic	Claude 3.5 Sonnet, Claude 3 Haiku
5	Ollama	Local models, fully offline
6	Custom	Any OpenAI-compatible endpoint

---

Security Model

• Single authorized user — only your Telegram user ID can send commands
• Encrypted config — API keys and tokens stored with Fernet encryption
• Tool sandboxing — AI can only invoke registered tools; no arbitrary code execution
• Injection protection — pattern-based prompt injection detection
• Path/command blocklist — system-critical paths and destructive commands blocked
• No inbound ports — only outbound connections to Telegram API
• Audit trail — every action logged to SQLite

---

Plugin API

Create a .py file in ~/.config/doit/plugins/:

PLUGIN_NAME = "my_plugin"
PLUGIN_VERSION = "1.0.0"
PLUGIN_DOIT_MIN_VERSION = "1.0.0"
PLUGIN_DESCRIPTION = "What this plugin does"

async def my_tool(arg1: str, count: int = 1, **kwargs) -> dict:
    # Must return a dict
    return {"result": f"Did {arg1} x{count}", "success": True}

TOOLS = [
    ("my_tool", "Description of my tool", my_tool, False),  # (name, desc, fn, is_dangerous)
]

Restart Doit. Your tool is now available to the AI.

---

CLI Commands

doit init        # Run onboarding wizard
doit run         # Start agent (used by OS service)
doit update      # Check and install updates
doit status      # Check if service is running
doit uninstall   # Remove service and optionally data

---

Resilience

Scenario	Behavior
Process crash	OS service auto-restarts; pending tasks recovered from DB
Network loss	Exponential backoff retry; never exits
Power loss	OS auto-start re-launches; state fully restored from SQLite
SIGTERM	Finish running tasks, persist state, close DB cleanly
AI unavailable	Notify user, retry with backoff, use fallback model
Resource overload	Delay task execution until CPU/RAM within limits

---

Database Schema

SQLite at ~/.config/doit/data/doit.db:

• tasks — all task executions with status, payload, result, retry count
• schedules — persistent schedules with next_run computation
• logs — structured application logs
• audit_log — tamper-evident audit trail
• config — runtime configuration
• plugin_registry — installed plugins
• migrations — schema version history

---

Requirements

• Python 3.10+
• A Telegram bot token (free, from @BotFather)
• An AI API key (NVIDIA, Zhipu, OpenAI, Anthropic) OR Ollama running locally

---

License

MIT — free to use, modify, and distribute.