Local AI.
No compromises.

The desktop inference engine that runs LLMs on your hardware. OpenAI-compatible API. MCP tools. Ollama model reuse. Free forever.

Download for Free View Source
$ npm i -g darksol
Darksol Studio — Desktop Application
$0.00
Cost per inference
69+
Tests passing
100%
Local & private
∞
Models supported

Everything you need. Nothing you don't.

Built for developers and power users who want full control over their AI stack.

⚡

Hardware-Aware Inference

Auto-detects GPU, VRAM, CPU cores, and RAM. Optimizes gpu_layers, threads, and context size automatically.

🔌

OpenAI-Compatible API

Drop-in replacement. /v1/chat/completions, /v1/completions, /v1/models, /v1/embeddings. SSE streaming built in.

🦙

Ollama Model Reuse

Already have Ollama models? Darksol finds and runs them directly — no re-download, no daemon required.

🔍

HuggingFace Directory

Browse, search, and pull GGUF models. Hardware-aware fit indicators tell you what runs before you download.

🔧

MCP Tool Integration

Connect external tools via Model Context Protocol. CoinGecko, DexScreener, Etherscan, DefiLlama — pre-configured.

🔑

API Keys & Gateway

Access large cloud models through Bankr LLM Gateway when you need them. Manage API keys securely from the app.

💰

Cost Tracking

Every local inference is $0.00. Track your usage, tokens processed, and savings vs cloud providers in real time.

🌡️

Thermal Monitoring

Real-time GPU/CPU temperature tracking. Know when your hardware is hot before it throttles your inference.

🛠️

Tool Use & Function Calling

Enable models to call functions, execute code, and access files. Configurable per-session from the app settings panel.

Download Darksol Studio

Free. No account required. Your data never leaves your machine.

🪟

Windows

Windows 10+ · x64 · Beta
Download Beta Installer (.exe) · ~90 MB
🍎

macOS

macOS 12+ · Intel & Apple Silicon
Coming Soon Universal .dmg · ~90 MB
🐧

CLI (npm)

Node.js 18+ · All platforms
Install via npm npm i -g darksol

Darksol vs Ollama

Same goal, different approach.

Darksol Studio

  • Hardware-optimized from birth
  • Reuses existing Ollama models
  • HuggingFace directory built in
  • MCP tool integration
  • Cost tracking & savings
  • Desktop app + web UI + CLI
  • Cloud gateway fallback (Bankr)
  • "Will it fit?" before download
  • Function calling & code execution

Ollama

  • Basic hardware detection
  • Own model format only
  • Manual model discovery
  • No tool integration
  • No cost awareness
  • CLI only
  • Local only, no fallback
  • Download first, find out after
  • No function calling

Or just use the CLI

Everything works from the command line too.

# Install globally
$ npm i -g darksol

# Search models (with hardware fit check)
$ darksol search llama
  llama-3.2-3b-gguf   3.2B  Q4_K_M  ✅ will fit
  llama-3.1-70b-gguf  70B   Q4_K_M  ❌ won't fit

# Pull and run
$ darksol pull llama-3.2-3b-gguf
  Downloading... 100% (1.8 GB) ████████████████████ done

$ darksol run llama-3.2-3b "explain quantum computing"
  Quantum computing uses qubits instead of classical bits...

# Use existing Ollama models directly
$ darksol run ollama/llama3.2 "hello world"
  Hello! How can I help you today?

# Start the API server
$ darksol serve
  Server started at http://127.0.0.1:11435

Your models. Your hardware.
Your rules.

No cloud. No subscriptions. No data leaving your machine.

Download Darksol Studio View Source on GitLab