🦁 LionLLM / LionAI v1.0

A fully modular, production-ready cognitive AI assistant built natively in Rust. LionLLM wraps local large language models (via Ollama) with a biology-inspired cognitive graph, a persistent semantic memory store, multimodal perception channels (vision and audio), a 7-stage reasoning pipeline, and an autonomous ReAct-based agent loop with local tool access.

---

🏗 System Architecture

The project is structured as five specialized crates:

graph TD
    lion_cli[lion_cli - Main REPL] --> lion_senses[lion_senses - Multimodal]
    lion_cli --> lion_agent[lion_agent - ReAct Loop & Tools]
    lion_cli --> lion_brain[lion_brain - Reasoning & Context]
    
    lion_senses --> lion_core[lion_core - Embeddings & Knowledge]
    lion_agent --> lion_brain
    lion_brain --> lion_core

Loading

1. lion_core: Low-level sensory cortex containing the 1.58-bit Ternary Encoder, persistent Knowledge Graph, and Long-Term Memory.
2. lion_brain: Cognitive engine driving the 7-stage Thinking+ pipeline (Understand $\to$ Remember $\to$ Retrieve $\to$ Reason $\to$ Generate $\to$ Verify $\to$ Optimize), Ollama client, and token context manager.
3. lion_senses: Multimodal perception wrappers supporting 8x8 pixel downsampling, Sobel edge calculation, and WAV audio analysis.
4. lion_agent: Autonomous ReAct loop with local tools (calculator, file read/write, safe shell runner, and text-only web fetcher).
5. lion_cli: Main interactive terminal REPL with streaming response capabilities and custom slash commands.

---

🚀 Getting Started

1. Compile the Project

Build the workspace in release mode:

cargo build --release

The compiled binary will be located at target/release/lion.

2. Run the Interactive Console

Run the unified CLI console directly:

cargo run --release -p lion_cli

3. Run the Integrated Tests

Verify all sixteen phases of implementation:

cargo test -p lion_cli

---

⚡ Slash Commands

Inside the REPL console, you can run commands directly using slash prefixes:

Command	Arguments	Description
/help	None	Displays help guidelines.
/status	None	Displays system, memory, and Ollama connectivity status.
/memory	None	Shows total persistent memory entries loaded.
/clear	None	Flushes the current conversation history.
/tools	None	Lists all tools available to the ReAct agent.
/use	<tool> <input>	Direct tool execution bypassing the LLM.
/calc	<expression>	Computes math expressions instantly using native Rust.
/image	<path>	Encoders an image into a 32-dim vector and generates an LLM description.
/audio	<path.wav>	Analyzes key features (RMS, frequency bands, ZCR) of a WAV file.
/agent	<task>	Launches the autonomous ReAct tool loop to complete a complex task.
/save	None	Forces memory serialization to disk.
/exit	None	Exits the program (automatically saving memory to ~/.lionai/memory.bin).

---

🔬 Local LLM Setup (Optional)

To unlock full generative text responses and visual descriptions:

1. Download and run Ollama.
2. Pull the default language model:

   ollama pull gemma3:1b

3. (Optional) Pull the vision model:

   ollama pull moondream

If Ollama is offline, LionLLM degrades gracefully into local memory-driven fallback mode and direct tool execution.