# Ollama (local runtime)

Local model runner and HTTP API, scoped with cloud features disabled.

Revision: 2026-09-15.1 · Review: unreviewed

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

## Applicability

Repository revision a43fad18b088095de20fbd7a8f0de50824cf5d27; living documentation separately dated.. Local Ollama server with OLLAMA_NO_CLOUD=1 and preloaded local models; excludes Ollama Cloud and web search..

## Claims

- **yes · documented**: Runs local models with cloud features disabled. [Ollama FAQ](https://docs.ollama.com/faq) Scope: Set OLLAMA_NO_CLOUD=1 and restart; prefetch model files separately.
- **yes · documented**: Supports Apple GPU acceleration through Metal. [Ollama hardware support](https://docs.ollama.com/gpu) Scope: Supported Apple hardware; actual memory fit unmeasured.
- **yes · documented**: Documents CPU execution and GPU offload modes. [Ollama FAQ](https://docs.ollama.com/faq) Scope: Backend placement can be inspected using ollama ps.
- **yes · documented**: Supports NVIDIA GPUs within documented driver and compute requirements. [Ollama hardware support](https://docs.ollama.com/gpu) Scope: Check the hardware compatibility list for the chosen GPU.
- **yes · documented**: Imports local GGUF model files through a Modelfile. [Ollama model import](https://docs.ollama.com/import) Scope: Supported model architectures; import does not quantize the GGUF file.
- **yes · documented**: Can load previously quantized GGUF models. [Ollama model import](https://docs.ollama.com/import) Scope: Prepare quantization before GGUF import; no quality equivalence asserted.
- **yes · documented**: Exposes a local HTTP API. [Ollama FAQ](https://docs.ollama.com/faq) Scope: Default bind is 127.0.0.1:11434; changing it affects network exposure.
- **yes · documented**: Supports a subset of OpenAI-compatible API fields and endpoints. [Ollama OpenAI compatibility](https://docs.ollama.com/api/openai-compatibility) Scope: Only the documented endpoint and parameter subset.
- **yes · documented**: Accepts JSON schemas for structured generation. [Ollama structured outputs](https://docs.ollama.com/capabilities/structured-outputs) Scope: Local runtime only; cited documentation excludes Ollama Cloud support.
- **yes · documented**: Streams generated output as newline-delimited JSON. [Ollama API streaming](https://docs.ollama.com/api/streaming) Scope: Native API streaming protocol; separate from compatible API streaming.
- **yes · documented**: Can process multiple requests to the same model in parallel. [Ollama FAQ](https://docs.ollama.com/faq) Scope: Requires sufficient memory; parallel request count multiplies context memory and excess work queues.

## 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?

## 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.