# LM Studio (local runtime)

Desktop runtime management and local serving through llama.cpp or MLX engines.

Revision: 2026-09-15.1 · Review: unreviewed

Commissioned by a Nimbalyst maintainer. Documentation synthesis; no independent review or product execution.

## Applicability

Living documentation accessed 2026-09-15; app and installed engine versions are not pinned.. Local LM Studio app and server; pin the separately installed llama.cpp or MLX engine. Excludes remote hosts, Bionic workflows and external tools..

## Claims

- **yes · documented**: Local chat and server operation can work offline. [LM Studio offline operation](https://lmstudio.ai/docs/app/offline) Scope: Download models and runtimes first; discovery, downloads and app updates need connectivity.
- **yes · documented**: Supports Apple Silicon Macs and the MLX engine. [LM Studio app documentation](https://lmstudio.ai/docs/app) Scope: Check the documented macOS and hardware requirements; Intel Mac support is not implied.
- **yes · documented**: Runs supported GGUF models with its llama.cpp engine. [LM Studio app documentation](https://lmstudio.ai/docs/app) Scope: Engine version is independently managed; supported model architectures only.
- **yes · documented**: Runs supported MLX models on Apple Silicon. [LM Studio app documentation](https://lmstudio.ai/docs/app) Scope: MLX engine only; does not imply GGUF and MLX artifacts are interchangeable.
- **yes · documented**: Provides a local inference server. [LM Studio offline operation](https://lmstudio.ai/docs/app/offline) Scope: Local inference with preloaded models; remote tools are excluded.
- **yes · documented**: Exposes an OpenAI-style chat-completions endpoint. [LM Studio chat-completions endpoint](https://lmstudio.ai/docs/developer/openai-compat/chat-completions) Scope: Documented request fields only; no complete OpenAI API parity claim.
- **yes · documented**: Supports streaming in the chat-completions API. [LM Studio chat-completions endpoint](https://lmstudio.ai/docs/developer/openai-compat/chat-completions) Scope: stream request parameter; compatible local model and engine.
- **yes · documented**: Accepts JSON schema constraints for generation. [LM Studio structured output](https://lmstudio.ai/docs/developer/openai-compat/structured-output) Scope: GGUF uses llama.cpp grammars and MLX uses Outlines; reliability varies by model.
- **yes · documented**: Provides draft-model speculative decoding. [LM Studio speculative decoding](https://lmstudio.ai/docs/app/advanced/speculative-decoding) Scope: Main and draft vocabularies must match; added resource use may slow generation.
- **partial · documented**: Documents parallel requests for the llama.cpp engine; MLX support is listed as forthcoming. [LM Studio parallel requests](https://lmstudio.ai/docs/app/advanced/parallel-requests) Scope: Partial across the target engines: llama.cpp continuous batching only at capture date; configure Max Concurrent Predictions.

## Open questions

- Which pinned model, quantization, context length and runtime versions fit the selected Mac?
- What are measured latency, throughput, peak memory and swap under one and four concurrent requests?
- The schema has no proprietary source-access value: Not established does not override the explicitly recorded proprietary app terms.

## Verification plan

1. Choose hardware and obtain explicit approval for installations, downloads and execution resources.
2. Pin runtime and engine versions, model artifact hashes, tokenizer, template, quantization and inference settings.
3. Run the local-model-mac investigation with synthetic prompts and capture raw measurements; do not record documentation as execution.

Last checked: 2026-09-15. Recheck by: 2026-10-15.

Content rights: Unlicensed prototype; redistribution decision pending.