O narzędziu
LM Studio — free desktop app do download i run AI modeli lokalnie. Data prywatne (local interactions). lmster — LM Studio core bez GUI dla Linux/cloud/CI deployment. Local APIs i SDKs: TypeScript, Python, REST. OpenAI i Anthropic-compatible endpoints. Native v1 REST API od LM Studio 0.4.0. Modele: GPT-OSS, Qwen3, Gemma3, LLaMA, Mistral, DeepSeek, Phi. LM Studio Pro opcjonalnie (early access + priority support).
Funkcje główne
- •Free desktop app (Windows, macOS, Linux).
- •Local LLM running (private data).
- •Lmster — LM Studio core bez GUI (headless deployment).
- •Linux boxes, cloud servers, CI integration.
Funkcje dodatkowe
▶Free Desktop App
Bezplatna aplikacja desktopowa dla Windows, macOS i Linux — najprostsza droga do local LLM. GUI-first podejscie dla non-technical users, bez wymaganej znajomosci terminal/CLI.
▶Local LLM Running
Pelne uruchamianie modeli lokalnie z private data — zero wysylania na zewnatrz. Krytyczne dla pracy z confidential documents, healthcare data i regulated industries.
▶lmster Headless
lmster — LM Studio core bez GUI dla headless deployments. Linux boxes, cloud servers, CI integration. Pozwala uzywac LM Studio engine w production bez desktop dependency.
▶Native v1 REST API
Native v1 REST API od LM Studio 0.4.0 — stable production API bez wrapping. Pozwala na integration z aplikacjami bez konieczno ranego mock'owania OpenAI endpoint.
▶Python SDK + TypeScript SDK
Official SDKs: lmstudio-python i lmstudio-typescript. Idiomatic API w obu jezykach, lepsze niz proste REST wrappery — error handling, streaming, type safety.
▶OpenAI + Anthropic-compatible Endpoints
OpenAI-compatible endpoints (chat.completions, embeddings) plus Anthropic-compatible (Messages API). Drop-in replacement dla obu major API patterns w branzy.
▶Built-in Chat Interface
Wbudowany chat UI z conversation history, system prompts, parameter tuning (temperature, top-p, max tokens). Pelnoprawny ChatGPT-like interface dla lokalnych modeli.
▶Model Browser z HF
Wbudowana przegladarka modeli z Hugging Face — search, filter, download bezposrednio z UI. Quantization options (Q4_K_M, Q5, Q8) automatycznie matched do hardware.
▶RAG Support
Chat z dokumentami — upload PDF/MD/TXT, LM Studio robi embeddings i retrieval. Built-in RAG bez external vector database — perfect dla solo developers i privacy-first workflows.
▶Multi-model Serving + GPU Offloading
Jednoczesne running wielu modeli z GPU offloading (CUDA, Metal, ROCm). Mozliwosc fine-tuning ile warstw modelu trafia na GPU vs CPU dla optymalnej performance per hardware.
✓ Zalety
Modele
- •GPT-OSS, Qwen3, Gemma3 (najnowsze).
- •LLaMA (3, 4), Mistral, DeepSeek, Qwen, Phi.
- •Hugging Face GGUF models.
- •Quantization options (Q4_K_M, Q5, Q8 itp.).
Cennik
- •Free: pełne features.
- •No usage fees.
- •LM Studio Pro: opcjonalny (early access builds + priority support).
- •Lmster: open-source dla deployment.
API i SDK
- •Native v1 REST API (od LM Studio 0.4.0).
- •Python SDK (lmstudio-python).
- •TypeScript SDK.
- •OpenAI-compatible endpoints.
- •Anthropic-compatible endpoints.
- •Drop-in replacement dla popular SDK.