Source Code: src/gaia/

Currently, the following capabilities are available, more will be added in the near future:

Platform Support Overview

Windows 11: Full GUI and CLI support with all features
Linux (Ubuntu/Debian): Full GUI and CLI support with all features

Use-Case Example	Function	Description	Platform Support
LLM Direct	Direct LLM queries via CLI	Direct model interaction using the new `gaia llm` command	Windows, Linux
Chat Agent	Interactive conversations with history	Interactive chat sessions with conversation context and commands	Windows, Linux
Code Agent	AI-powered code generation and analysis	Autonomous workflow for code generation, project scaffolding, and error fixing	Windows, Linux
Blender Agent	3D content creation and manipulation	Specialized agent for Blender automation and workflow	Windows, Linux
Summarization	Document and transcript summarization	AI-powered summarization with multiple output formats	Windows, Linux
Evaluation Suite	Model evaluation and benchmarking	Scenario-based agent eval, throughput benchmarking, and reporting	Windows, Linux
Voice Interaction	Speech-to-speech conversation	Voice-based AI interaction with TTS and ASR	Windows, Linux
Web Dashboards	Browser-based interfaces for agents	JavaScript apps with simple configuration	Windows, Linux

LLM Direct Usage

The gaia llm command provides direct access to language models without requiring server setup. This is the simplest way to interact with AI models:

# Basic query
gaia llm "What is 1+1?"

# Specify model and token limit
gaia llm "Explain quantum computing" --model Qwen3-0.6B-GGUF --max-tokens 200

# Disable streaming for batch processing
gaia llm "Write a short poem" --no-stream

Requirements: Requires lemonade-server to be running. Run gaia init to automatically install Lemonade Server and download models, or install manually from lemonade-server.ai. The command will provide helpful error messages if the server is not accessible. Platform Availability: Windows and Linux

Chat Agent

The Chat agent provides an interactive conversational interface with conversation history and various utility commands:

# Start interactive chat session
gaia chat

# Send a single message
gaia chat "What is machine learning?"

# Use a specific model
gaia chat --model Llama-3.2-1B-Instruct-Hybrid

# Use document Q&A with RAG (single file)
gaia chat --index manual.pdf

# Multiple documents
gaia chat --index doc1.pdf doc2.pdf

# Watch a folder and auto-index any PDF dropped into it
gaia chat --watch ./docs

# Show detailed statistics
gaia chat --show-stats

Key features:

Conversation History: Maintains context across messages with configurable history length
Document Q&A: RAG support for querying PDF documents with --index (one-off) or --watch (auto-index a folder)
Interactive Commands: Built-in commands for session management and debugging
Streaming Responses: Real-time response streaming for better user experience
Model Flexibility: Support for different LLM models with automatic prompt formatting
Single Message Mode: Non-interactive mode for scripting and automation

Interactive Commands:

/clear - Clear conversation history
/history - Show conversation history
/system - Show current system prompt configuration
/model - Show current model information
/prompt - Show complete formatted prompt sent to LLM
/stats - Show performance statistics (tokens/sec, latency, etc.)
/help - Show available commands
quit, exit, or bye - End the chat session

Requirements: Requires lemonade-server to be running. The chat agent defaults to the Gemma-4-E4B-it-GGUF model. Platform Availability: Windows and Linux

Blender Agent

The Blender agent provides specialized functionality for 3D content creation and workflow automation. It is now fully integrated into the CLI for easy access:

# Run all Blender examples
gaia blender

# Interactive 3D scene creation
gaia blender --interactive

# Create specific objects with custom queries
gaia blender --query "Create a red cube and blue sphere"

# Run a specific example
gaia blender --example 2

Key capabilities:

Scene Management: Clear scenes and get scene information
Object Creation: Create cubes, spheres, cylinders, cones, and torus objects
Material Assignment: Set RGBA colors and materials for objects
Object Modification: Modify position, rotation, and scale of existing objects
Interactive Planning: Multi-step scene creation with automatic planning

Requirements: In addition to the Lemonade server, the Blender agent requires a Blender MCP server to be running. See the CLI documentation for complete setup instructions. Platform Availability: Windows and Linux

Summarization

The Summarization feature provides AI-powered document and transcript processing with multiple output formats:

# Summarize a meeting transcript
gaia summarize -i meeting_transcript.txt -o summary.json

# Summarize emails with PDF output
gaia summarize -i emails_directory/ -f pdf -o summaries/

# Generate specific summary styles
gaia summarize -i document.txt --styles executive action_items --max-tokens 1024

# List available configuration templates
gaia summarize --list-configs

Key features:

Multiple Input Types: Process transcripts, emails, and documents
Flexible Output Formats: JSON, PDF, email, or both
Configurable Styles: Brief, detailed, bullets, executive, participants, and action items
Batch Processing: Process entire directories of documents
Template Support: Use predefined configuration templates
Model Flexibility: Support for both local (Lemonade) and cloud (OpenAI/Claude) models

Output formats:

json - Structured JSON with all summary components
pdf - Professional PDF reports
email - Email-ready format with recipients
both - Generate both JSON and PDF

Summary styles (--styles ..., space-separated):

brief - Concise 2-3 sentence summary
detailed - Comprehensive detailed summary
bullets - Bullet-point format focused on action items
executive - High-level executive summary
participants - Meeting participants and roles
action_items - Extracted action items with owners/deadlines
all - Generate every style above

Platform Availability: Windows and Linux

Evaluation Framework

GAIA includes a comprehensive evaluation framework for testing and comparing AI model performance:

# Run the agent eval benchmark (scenario-based, end-to-end)
gaia eval agent                                 # all scenarios
gaia eval agent --category rag_quality          # one category
gaia eval agent --scenario simple_factual_rag   # one scenario

# Compare a run against a saved baseline
gaia eval agent --compare eval/results/latest/scorecard.json

# Email-triage throughput benchmark (tokens/sec, TTFT, latency)
gaia eval benchmark --model Gemma-4-E4B-it-GGUF --limit 50

# Generate a report from an evaluation results directory
gaia report -d ./output/evaluations -o report.md

# Visualize llama.cpp server performance logs
gaia perf-vis ./logs/llama-server.log --show

Key capabilities:

Scenario-Based Agent Eval: Multi-turn conversations judged across scoring dimensions
Throughput Benchmarking: Measure tokens/sec, time-to-first-token, and pipeline latency
Baseline Comparison: Diff a run against a saved scorecard to catch regressions
Auto-Fix Mode: --fix invokes Claude Code to repair failures and re-evaluate
Report Generation: Markdown/JSON/JUnit output for CI integration
Performance Visualization: Plot llama.cpp server metrics from log files

Platform Availability: Windows and Linux

Voice Interaction (Talk)

The Talk feature enables voice-based conversations with AI models:

# Start voice conversation
gaia talk

# Voice chat with document Q&A (RAG)
gaia talk --index manual.pdf

# Use specific model with voice and document
gaia talk --model Qwen3-0.6B-GGUF --index report.pdf

# Disable text-to-speech (ASR only) with document
gaia talk --no-tts --index technical_manual.pdf

# Configure audio settings with document
gaia talk --audio-device-index 1 --whisper-model-size medium --index doc.pdf

Key features:

Speech Recognition: Whisper ASR for voice input
Text-to-Speech: Kokoro TTS for natural voice output
Document Q&A: RAG support for hands-free document queries
Real-time Processing: Streaming audio pipeline
Device Selection: Configure audio input devices
Model Flexibility: Choose ASR model sizes

Platform Availability: Windows and Linux

Desktop Applications

GAIA provides desktop applications with graphical interfaces for interacting with AI capabilities. These applications are built with Electron and provide native desktop experiences. For available applications and setup instructions, see the Desktop UI documentation. Platform Availability: Windows and Linux

Supported LLMs

GAIA runs models locally through Lemonade Server, so any model in the Lemonade catalog is supported — GGUF models run via llama.cpp on GPU/CPU, and FLM models run on the Ryzen AI NPU (XDNA2). To request support for a new model, contact the AMD GAIA team. The models GAIA agents use by default (pulled by gaia init --profile <profile>):

Model	Format / Backend	Used by
`Gemma-4-E4B-it-GGUF`	GGUF (llama.cpp)	Default chat, coding, and vision (multimodal)
`Qwen3.5-35B-A3B-GGUF`	GGUF (llama.cpp)	Specialized agents (analyst, browser, file I/O, code, jira, docker, routing, doc Q&A)
`Qwen3-4B-Instruct-2507-GGUF`	GGUF (llama.cpp)	Summarization
`Qwen3-VL-4B-Instruct-GGUF`	GGUF (llama.cpp)	Alternate vision model
`embeddinggemma-300m-GGUF`	GGUF (llama.cpp)	Embeddings for RAG / document Q&A
`gemma4-it-e2b-FLM`	FLM (NPU)	Chat on Ryzen AI NPU (`--profile npu`)
`embed-gemma-300m-FLM`	FLM (NPU)	Embeddings on Ryzen AI NPU (`--profile npu`)
`SDXL-Turbo`	Stable Diffusion	Image generation (also `SDXL-Base-1.0`, `SD-Turbo`, `SD-1.5`)

Installing Additional Models

GAIA uses Lemonade Server for model management. Models can be installed through:

GAIA CLI: Run gaia init --profile <profile> to download models for a specific use case. Valid profiles: minimal, sd, chat, code, rag, mcp, vlm, email, npu, all.
System Tray Icon: Access the Lemonade model manager from the system tray
Web UI: Manage models through the Lemonade web interface

To remove all downloaded models: gaia uninstall --purge --purge-models --yes (add --purge-hf-cache to also clear ~/.cache/huggingface/hub). For detailed instructions on installing and managing models, see the Lemonade Server documentation or the CLI Reference.

Command Line

API & Development

Help

Features

Platform Support Overview

LLM Direct Usage

Chat Agent

Blender Agent

Summarization

Evaluation Framework

Voice Interaction (Talk)

Desktop Applications

Supported LLMs

Installing Additional Models

License

​Platform Support Overview

​LLM Direct Usage

​Chat Agent

​Blender Agent

​Summarization

​Evaluation Framework

​Voice Interaction (Talk)

​Desktop Applications

​Supported LLMs

​Installing Additional Models

​License

Platform Support Overview

LLM Direct Usage

Chat Agent

Blender Agent

Summarization

Evaluation Framework

Voice Interaction (Talk)

Desktop Applications

Supported LLMs

Installing Additional Models

License