Models & BYOK

Bring your own key

MythDesk works with:

• OpenAI-compatible endpoints
• Anthropic
• Gemini
• OpenRouter

Keys are stored in Windows Credential Manager:

mythd secret set GEMINI_API_KEY

Run local models (free, private)

Ollama, LM Studio and llama.cpp servers are supported.

ollama pull qwen2.5:14b          # or gemma3:27b if you have ~24 GB VRAM
mythd local detect               # finds Ollama, LM Studio and llama.cpp servers
mythd local setup                # adds a local model per role as a free fallback
mythd local setup --all-local    # never call the cloud

To check which roles a local model can hold, the project includes a benchmark script:

python examples/bench_local.py --provider ollama --model gemma3:27b

Roles

Pick a model per role:

Router

The model router handles fallback between models, keeps a cost log and enforces a budget cap. The desktop app shows this month's spend.