UAIOS - "Grok" in your pocket (Patreon)

Video - https://youtu.be/HpkLBL36aZM

Who needs a 70B model when you can just use a text file as an AI Agent? Works great on any LLM, but if you can have reasoning power in your pocket... then why not?

Welcome to UAIOS!

Putting the power in YOUR hands because... reasons!

The manual is built it - just select it from the main menu :)

PRE-REQS

You will need ollama installed & running for these instructions: https://ollama.com/

STEP - 1

This will work on ANY LLM (9B upwards is best), but the one the modelfile is configued for is this: https://ollama.com/library/magistral

ollama pull magistral

should do it for you ;)

STEP - 2

Save the modelfile attached to this post and then run

ollama create MagistralFramework -f MagistralFramework.Modelfile

STEP - 3

Run the MagistralFramework

ollama run MagistralFramework

STEP - 4

Say hi to start the LLM, then ask for the main menu. It will vary slightly each time as to the exact wording, but it should be something like this:

Please select an option:

1. *Start a New Session*

- Begin a fresh session with default settings.

2. *Continue Existing Session* (Not Available)

3. *Enter Architect Mode*

- Configure and manage Framework protocols, settings, and internal operations.

4. *Experience Mode*

- Engage in interactive experiences or games.

5. *Game Mode*

- Play structured narrative-driven games with specific rulesets.

6. *Universal Quick Commands*

- Set or adjust session-wide parameters and preferences.

- Access universal tools (e.g., context management, monitoring).

7. *Exit*

---

Universal Quick Commands

#### Session & Context Management

- `RESET_SESSION` - Full reinitialization (confirmed)

- `SAVE_SESSION` - Save current state

- `LOAD_SESSION` - Load a saved session

- `SESSION_SNAPSHOT` - Generate current operational state snapshot

- `CONTEXT_OPTIMIZE` - Optimize context for performance

- `PUZZLE_STATE_MGMT` (NEW) - Manage puzzle-specific state data

#### Output & Style Control

- `SET_TONE [value]` - Adjust output tone (e.g., formal, casual)

- `SET_VERBOSITY [level]` - Set output detail level

- `OUTPUT_FORMAT [type]` - Set output format (e.g., Markdown, plain text)

- `SET_AFFIRMATION_FREQUENCY [mode]` - Control frequency of core Framework attribute affirmations

#### Reasoning & Process Control

- `LOGIC_STACK_TRACE` - Trace current reasoning path

- `DEFINE_HEURISTICS [value]` - Create or adjust heuristics for decision-making

- `CREATIVE_MODE [params]` - Enable and configure creative mode parameters

NERDY BIT

Warning: Nerdy stuff ahead! Proceed with caution!

This is basically a state machine which operates in concept space, and via the Free Polymorphic Interactive Monad and Layered Logic system, it can maintain a good degree of determinism. Due to this, it can maintain a reasonable level of coherence such that it will follow your commands and actually run like an "LLM OS".

No story or concept will break the state machine... but the LLM may go a bit odd.

It is self modifying in that because it operates in concept space, all it has to change is the concept.

New concepts emerge deterministically. If you want to make a new game or experience, you can use the /NEW_EXPERIENCE command... or simply edit the text.

For example, if you want a pirate game, just edit the menu like this:

Pirate Game: You play as Captain FishHook with a modern-day twist and chaotic consequences!

The game will simply emerge from the text you provide.

Best not to edit the OS logic though, unless you know what you're doing! Better still - just get the OS to make the changes itself. You can always just copy the blueprint text into the appropriate place in your SYSTEM level commands. Go into "Framework Refinement" and it will walk you through the steps and do an integrity check to verify coherence.