Solaria Agent Documentation

Everything you need to set up, configure, and master Solaria Agent — the local, private AI assistant for your desktop.

System Requirements

Solaria Agent is built with Tauri 2 and can run on Linux, with Windows and macOS support planned.

Minimum Requirements:

★Node.js ≥ 18 (LTS recommended)
▪Rust (via rustup) – for Tauri backend
▪Tauri CLI (cargo install tauri-cli)
▪Linux: libwebkit2gtk-4.1, libappindicator3, libgtk-3, libsoup-3.0
▪4 GB RAM minimum (8 GB recommended for local models)
▪Git (for cloning and updates)

Installation

Method 1: One-liner (Recommended)

curl -fsSL https://raw.githubusercontent.com/Angelcmp/solaria/main/install.sh | bash

Method 2: Build from Source

git clone https://github.com/Angelcmp/solaria.git
cd solaria
npm install
npm run tauri dev

For production builds, use npm run tauri build. The binary will be in src-tauri/target/release/.

Quick Start

After launching Solaria Agent, here is how to get productive quickly:

  1. Install Ollama and pull your first model: ollama pull qwen3.5
  2. Open Settings (gear icon) → AI Providers → select your provider
  3. For cloud providers, paste your API key in the settings panel
  4. Start chatting! Type a message and press Enter
  5. Try Agent Mode: click "[🤖 AGENT]" and give it a task like "check my git status"

AI Providers

Local (Ollama)

Run models entirely offline on your machine. No internet needed for inference.

Qwen3.5Llama 3.2Llama 3.1DeepSeek-R1MistralGemma 3Gemma 4Phi-3CodeGemmaCodeQwen

BYOK Cloud

Bring your own API keys. Data is sent to the provider but Solaria never stores or logs it.

OpenAI
GPT-4o, GPT-5, o1, o3
Anthropic
Claude 3.5, Claude 4
Google
Gemini 2.0, Gemini 2.5
Groq
Llama 3.3, Mixtral
DeepSeek
DeepSeek-V3, R1
Cohere
Command R+, Aya
Moonshot
Kimi K2
Z.ai (GLM)
GLM-4

Agent Mode

Agent Mode gives Solaria real access to your system. When activated, the AI can plan and execute multi-step tasks using tools. Every action requires explicit confirmation before execution.

⚠️ Security Notice

Agent Mode runs commands on your real system. Always review commands before confirming. Use Docker or a VM for untrusted tasks.

Activate by clicking the [🤖 AGENT] button in the chat interface. The agent will:

  • Plan steps to complete your task
  • Execute tools one at a time, showing results
  • Adapt based on intermediate results
  • Provide a final summary with all findings

Tools Reference

shellExecute terminal commands
Examples: ls -la, git status, npm install, cargo build, docker ps, cat file.txt
read_fileRead file contents from disk
Examples: /var/log/syslog, src/main.rs, package.json, .env.example
write_fileWrite or modify files on disk
Examples: Create config.toml, edit .gitignore, write helper.sh
globFind files matching a pattern
Examples: **/*.ts, src/**/*.rs, **/*.test.tsx, apps/*/package.json
grepSearch text content inside files
Examples: TODO, FIXME, export default, fn main, async function

Skills System

Skills are predefined workflows that guide the AI through specific tasks. Each skill includes a system prompt, tool access rules, and output formatting.

Built-in Skills

deep-researchmarket-analysisreport-generatormeeting-notesdata-analysiscode-reviewtranscription-processingknowledge-builder

Skill Factory

Create custom skills directly from the chat. Describe what you want the skill to do and Solaria generates the configuration. Skills are stored locally and can be exported/imported via JSON.

Memory & RAG

Solaria Agent features a persistent memory system powered by RAG (Retrieval-Augmented Generation). It uses vector embeddings stored in SQLite with the sqlite-vec extension.

How it works:

  1. Conversations and project files are automatically indexed into vector embeddings
  2. When you ask a question, relevant past context is retrieved via semantic search
  3. The relevant context is injected into the system prompt for the AI
  4. All embeddings are stored locally in ~/.solaria/memory.db

Configuration Options

Top-KNumber of results to retrieve5
Score ThresholdMinimum relevance score0.5
Chunk SizeText segment size for indexing512
Chunk OverlapOverlap between adjacent chunks128
Recency DecayWeight decay for older content0.95

Projects

Organize your work with the Projects system. Each project has its own conversation history, indexed files, and skill set. Projects appear in the sidebar with full CRUD operations.

CLI Commands

Solaria Agent includes a CLI for automation, scripting, and headless operation.

solaria ask "your question"Ask a single question via CLI
solaria agent "task description"Run a task in agent mode
solaria serveStart in headless server mode
solaria memory index ./srcIndex project files for RAG
solaria memory search "query"Search the memory store
solaria memory statsView memory storage statistics

Configuration

All settings are stored locally. Access them via the gear icon in the app. API keys are stored securely using the OS keyring (no plaintext on disk).

Settings include: AI provider selection, API keys (per provider), model selection, temperature, max tokens, Ollama endpoint URL, web search (Tavily) API key, memory/RAG parameters (top-K, threshold, chunk size), and UI preferences.

Security

Solaria Agent is designed with security in mind. Here is the security model:

🔒 Prompt Injection Protection
Input is sanitized and validated before reaching AI models.
🛡️ Shell Injection Prevention
Commands in agent mode are escaped and validated. Dangerous patterns are blocked.
🔐 API Key Storage
Keys are stored in the OS keyring (Keychain on macOS, Secret Service on Linux). No plaintext on disk.
🚫 No Telemetry
Zero telemetry, analytics, or tracking of any kind. We cannot see your data even if we wanted to.
🌐 Content Security Policy
CSP headers prevent XSS via rendered markdown. Only safe markdown rendering is used.
🛤️ Path Traversal Protection
File operations are restricted to authorized directories. Path traversal attempts are blocked.

Troubleshooting

Ollama models not appearing?
Make sure Ollama is running (ollama serve) and you have pulled at least one model (ollama pull qwen3.5). Check the Ollama endpoint URL in settings.
API key not working?
Verify the key in the provider dashboard. Keys are stored in the OS keyring — any app with keyring access can conflict. Try removing and re-adding the key.
App won't start on Linux?
Install missing dependencies: sudo apt install libwebkit2gtk-4.1-dev build-essential curl wget file libxdo-dev libssl-dev libayatana-appindicator3-dev librsvg2-dev
Memory/RAG not returning results?
Check that index_project_files has been called. Lower the score threshold in settings. Verify ~/.solaria/memory.db exists and is not corrupted.
Agent Mode stuck or slow?
The agent may be waiting for a confirmation. Check the side panel for a pending confirmation. Complex tasks can take 30+ seconds. Reduce task scope or use a faster model.