An open-source AI system built to reduce hallucinations, run any model, and handle long autonomous tasks without babysitting the conversation.
Whether "What's running at home?" or "Show me all active containers" — for TRION both resolve to the exact same predicate and follow the same pipeline route.
TRION decides upfront which operation is permitted. Layers further down the pipeline cannot override this contract on their own.
Deterministic pre-sorting keeps resource usage low. Only complex tasks activate expensive LLMs — everyday questions are handled locally at full speed.
try / except ValueError → log + reraiseTRION exposes its planning pipeline in every response. You see the exact route the request took, how many steps were planned, which tools ran — and whether the Verifier issued anapproved verdict before any output reached you.
No guess, no silent retry. If the plan is rejected, the trace says so.
Full architecture on the system map→
Every MCP, plugin, and tool you install appears in the Launchpad automatically — no manual registration, no configuration file to edit. TRION discovers it the moment it's installed.
Drag any app into the dock to keep it one click away. The screen stays clear. Your workspace stays yours. TRION doesn't feel like a chat interface — it feels like an operating system.
Drag any app icon from the Launchpad into the Dock to pin it. One click away, always — without touching a config file.
Drop a bundle, pick a file, or paste a GitHub URL. TRION sets up the runtime, registers the tools, and makes them available to the AI immediately — no config files, no restart.
Drop an archive, browse from disk, or paste a GitHub URL.
Paste any repository URL. Branch or tag is parsed from the path.
TRION has been tested and developed on macOS Tahoe 26.5.1 with the Apple M4 chip. Ubuntu support will follow.
TRION lets you use a cloud API key instead. This means you can run TRION on any machine — no local GPU required.
Nothing. TRION is model-agnostic. You can swap the model at any time. Memory, chats, tools, and MCPs stay in TRION — the flow works exactly as before.
Make sure Ollama is running on port 11434 — TRION finds it automatically on that port. If you want a custom port, update the docker-compose accordingly.
System prompts live in intelligence_modules/. They are fine-tuned for the pipeline. Modifying them can affect the entire pipeline — proceed at your own risk. Keep a copy of the folder before making changes.
Yes. TRION is free and open-source software under the AGPL-3.0 license — free to use, study, share, and modify, for individuals and companies alike. That was the whole point it was built for. Users first.
TRION is licensed under GNU AGPL-3.0, which already permits commercial use — provided you honor its copyleft terms (notably: if you run a modified version as a network service, you must make your source available under the same license). If your organization can't comply with the AGPL, a separate commercial license is available — get in touch at trlon.devs.dk@gmail.com.
You can support development via Stripe if you'd like — reach out and I'll send you a link. Every contribution helps keep TRION free and actively maintained.
Your data stays on your PC. No data is collected. The only exception: if you use a cloud model, your messages are shared with that API provider as usual — that's between you and them. If you use local models via Ollama, everything stays fully local. Memory is stored in a database on your own system, so make regular backups. There are no cloud backups.
Yes — if you use local models via Ollama, TRION works fully offline. Cloud model features require an internet connection to reach the respective API provider.
A direct API gives you a chat. TRION gives you a system. It drastically reduces hallucinations through a Verifier that has to explicitly approve every answer before it reaches you. It remembers what you tell it — that memory persists across every model you use. And it can run long, multi-step tasks autonomously while you do something else, with no need to babysit the conversation.
MCPs are tool connectors that give TRION access to external capabilities — file systems, browsers, calendars, custom services, and more. In TRION, MCPs are purpose-built for the system and install via the built-in UI. No manual configuration needed for most bundles.
Automatic tool discovery: When you install a new MCP, TRION picks up its tools immediately — no prompt update, no restart, no manual registration. The AI knows what the new tool can do the moment it's installed.
Built-in Cron server: TRION ships with a native scheduler. You can tell the AI to run a task on a schedule — every morning, every hour, once a week — and it executes autonomously while you're away. No third-party tool needed.
Clean uninstall, always: TRION tracks every file it creates per MCP. Removing one leaves nothing behind — no leftover registry entries, no orphaned files.
It's a planned maintenance mode, not a shipped feature yet. The idea: on a schedule or on request, TRION re-checks memory, re-evaluates outdated or unclear facts, and separates conversation claims from verified ones — without deleting anything, only hygienic upkeep. The architecture and rules are documented as a target design; implementation hasn't started.