Methodology

Local AI Stack is a documentation-backed beginner guide. It does not currently operate an independent benchmark lab, product test lab, privacy audit, security audit, or hardware certification program. The site can sound confident where official documentation supports a narrow claim, but it must not imply Local AI Stack independently reproduced that claim.

Public label	Meaning
Official documentation reviewed	A current official source, official download page, official repository, changelog, or platform/provider page was reviewed on the recorded date. This means the source was reviewed, not independently reproduced.
Official documentation reviewed, with caveats	Official documentation supports the narrow claim, but material limits, configuration dependencies, provider boundaries, or version-sensitive details remain.
Conservative estimate, not a benchmark	The guidance is a deterministic or editorial estimate based on public documentation, model-file concepts, memory concepts, context length, quantization, or runtime assumptions.
Third-party evidence	The claim is supported by authoritative documentation outside the product's own docs, such as Docker, Google Search Central, Hugging Face, OpenAI, or Anthropic documentation.
Not independently tested by Local AI Guide	Local AI Guide has not installed, run, benchmarked, audited, inspected, or validated the workflow. This can appear beside source-backed claims.
Unknown / do not rely on yet	The reviewed sources do not safely support the claim, or the claim needs manual source review or actual testing before publication.

A hands-on Local AI Stack test would need to record:

machine model or hardware class;
CPU, RAM, GPU, and dedicated VRAM if applicable;
operating system version;
app/runtime version;
model identifier and quantization;
context length;
install steps;
success state;
failure state if any;
memory, storage, and timing observations;
screenshots or logs tied to the exact test ID.

Without that packet, a page may be useful, but it is not an independently tested result.

The RAM/VRAM calculator is deterministic sizing guidance. It combines model size, quantization, context length, runtime overhead, workload overhead, and a safety margin.

Calculator outputs should always be read as:

Conservative estimate, not a benchmark.
Not lab measurements.
Not purchase guarantees.
Not proof that a model will run comfortably.

Actual results depend on model, quantization, context length, runtime, GPU offload, drivers, thermals, and other running apps. Local AI Guide has not independently benchmarked this hardware.

Privacy claims are workflow claims, not product slogans. Local AI Guide is documentation-backed, not a privacy audit. A local model, local app, local UI, and local-only workflow are different things.

Cloud providers, cloud embeddings, web search, telemetry, remote access, plugins, Docker/network exposure, and local-server configuration can change the privacy boundary. Local AI Guide has not independently audited network activity, telemetry, storage paths, or every provider configuration.

No fake benchmarks.
No fake screenshots.
No fake "tested by us" claims.
No fake affiliate rankings.
No unsupported "best" claims.
No unqualified claim that local AI is private.
No claim that every machine can run every model.

Methodology

Evidence model

Evidence labels

Calculator policy