triobot 0.2.2

trio.ai - train your own AI, deploy it everywhere

trio.ai

Your AI, everywhere.

trio.ai is an open-source AI agent framework. Train your own LLM, connect any provider (Ollama, OpenAI, Claude, Gemini), and deploy across CLI, Discord, Telegram, Slack, WhatsApp, Teams, and 16 more channels.

---

Install

pip install trio-ai
trio onboard

That's it. Two commands. trio onboard auto-detects Ollama, lets you browse 1,900+ skills, and sets everything up.

Then start chatting:

trio agent

From source

git clone https://github.com/iampopye/trio.git
cd trio
python install.py
trio onboard

Need Ollama? Install from ollama.com, then: ollama pull llama3.2:1b

---

What It Does

You type --> trio agent --> AI responds (streaming, with tools, memory, guardrails)

trio connects to any LLM and adds:

• 12 built-in tools: web search, browser, email, calendar, notes, shell, file ops, math, screenshot, RAG, URL reader, sub-agent delegation
• 1,900+ skills: coding, marketing, DevOps, security, finance, data science, creative writing, sysadmin, web dev
• Sub-agent system: delegate tasks to specialized agents (researcher, coder, reviewer, planner, summarizer)
• Persistent memory: remembers facts across conversations
• 5-layer guardrails: input/output filtering, jailbreak detection
• 17 chat channels: CLI, Discord, Telegram, Signal, Slack, WhatsApp, Teams, Google Chat, iMessage, Matrix, SMS, Instagram, Messenger, LINE, Reddit, Email
• TrioHub: community skill/plugin registry with 1,900+ skills across 13 categories
• Plugin system: extend with your own tools and skills
• Production daemon: runs as a system service with auto-restart and health monitoring
• Train your own model: the only open-source agent framework with built-in LLM training

---

Commands

trio onboard                # Interactive setup wizard (6 steps)
trio agent                  # Interactive chat
trio agent -m "hello"       # Single message
trio status                 # System overview
trio doctor                 # Diagnose issues
trio doctor --fix           # Auto-repair

trio gateway                # Start all chat channels
trio daemon start           # Run as background service
trio daemon install         # Auto-start on boot (systemd/launchd/Windows)
trio daemon status          # Health, uptime, channels

trio provider add           # Add LLM provider
trio pairing list           # Manage DM access
trio plugin list            # Manage plugins
trio skill list             # Browse installed skills
trio hub search "coding"    # Search TrioHub registry
trio skill install <name>   # Install from TrioHub

trio update                 # Self-update
trio train                  # Train your own model

---

Providers

Works with 13+ LLM providers out of the box:

Provider	Setup
Ollama (local)	ollama pull llama3.2:1b -- auto-detected
OpenAI	API key
Anthropic (Claude)	API key
Google (Gemini)	API key
Groq	API key
DeepSeek	API key
OpenRouter	API key
trio-max (built-in)	Train your own -- no API needed

---

Chat Channels

Deploy your AI on any platform:

Channel	Type
CLI	Built-in terminal chat
Discord	Bot with live message editing
Telegram	Bot with markdown support
Signal	Private messenger
Slack	Workspace bot (Socket Mode)
WhatsApp	Business API webhook
Teams	Bot Framework
Google Chat	Service account webhook
iMessage	macOS only (AppleScript)
Matrix	Element/Matrix rooms (matrix-nio)
SMS	Twilio API
Instagram	DM via Meta Graph API
Messenger	Facebook Messenger webhook
LINE	LINE Bot SDK
Reddit	Bot via PRAW
Email	IMAP receive / SMTP send

Enable channels: trio onboard or edit ~/.trio/config.json

---

Sub-Agents

trio can delegate tasks to specialized sub-agents:

Agent	Role	Tools
researcher	Web search, gather information	web_search, browser, RAG
coder	Write and debug code	shell, file_ops
reviewer	Review content for quality	none (LLM-only)
planner	Break tasks into steps	none (LLM-only)
summarizer	Condense long content	none (LLM-only)

The main agent automatically delegates when a task benefits from specialization.

---

TrioHub -- Community Registry

Browse and install skills from the community:

trio hub search "python"     # Search skills
trio hub trending            # Popular skills
trio skill install <name>    # Install a skill
trio skill list              # Your installed skills

13 categories: Coding (418), Web Dev (159), Marketing (163), Security (80), Finance (47), Data Science (102), Creative (79), SysAdmin (204), Education (25), Productivity (167), Legal (29), Health (21), General (415)

---

Architecture

User --> Channel --> MessageBus --> AgentLoop
                                      |
                                      +--> Build Context (memory + skills + tools)
                                      +--> Call LLM (Ollama / OpenAI / Claude / local)
                                      +--> Execute Tools (if needed)
                                      +--> Delegate to Sub-Agents (if needed)
                                      +--> Guardrails Filter
                                      |
                                   Response --> MessageBus --> Channel --> User

---

Project Structure

trio/
  trio/                     # Agent framework
    core/                   #   AgentLoop, MessageBus, Config, Memory, Sessions, SubAgents
    providers/              #   Ollama, OpenAI, Claude, Gemini, 10+ more
    channels/               #   17 channels (CLI, Discord, Telegram, Slack, WhatsApp, etc.)
    tools/                  #   12 tools (web, browser, email, calendar, notes, shell, etc.)
    skills/                 #   1,900+ markdown-based skills
    plugins/                #   Plugin system (loader, manager, manifest)
    hub/                    #   TrioHub community registry client
    cron/                   #   Daemon, heartbeat, scheduler
    shared/                 #   Guardrails, pairing security, context analyzer
    cli/                    #   13 CLI commands

  trio_model/               # LLM engine (train your own)
    model/                  #   Transformer (RoPE, GQA, RMSNorm, SwiGLU)
    training/               #   Pre-train, SFT, Constitutional AI
    inference/              #   FastAPI server + CLI

  triohub/                  # Community skill/plugin registry
    index.json              #   1,909 skills across 13 categories

---

Train Your Own Model

pip install trio-ai[model]
trio train                  # Start training (pause/resume with Ctrl+C)

Or on Kaggle/Colab with GPU:

# Upload notebooks/kaggle_train_trio.ipynb
# Uses T4 GPU with Flash Attention + gradient checkpointing

Config	Params	Hardware
trio-max	Auto	Any system -- auto-optimizes
nano	~1M	CPU (4GB+ RAM)
small	~125M	T4 GPU (16GB)
medium	~350M+	A100 / M2+

---

Platform Support

Platform	Status
Windows 10/11	Full support (NVIDIA CUDA)
macOS Intel	Full support
macOS Apple Silicon	Full support (MPS Metal)
Ubuntu / Debian	Full support (CUDA, ROCm)
Fedora / Arch	Full support
WSL2	Full support (CUDA passthrough)

---

License

MIT -- Karan Garg

---

Built from scratch. Train it. Deploy it. Own it.