About DoesItLocal

DoesItLocal is a living database that answers one question per task: can a local LLM safely do this, and if not, what should you use instead?

Most “which model” advice points you at a leaderboard. A benchmark rank tells you a model is generally strong — it does not tell you whether a 7B model running on your laptop will reliably extract the fields from your invoices, or quietly get one wrong. DoesItLocal scores the task, not the model: every task carries one current verdict, the local models that clear it, and a recommended fallback when none do.

Why local

• Privacy. Proprietary code and personal data never leave your machine. For a privacy-bound task, local is the safer route even when a cloud model scores higher.
• Cost. A task that runs safely on a model you already have is free. The product optimizes for the cheapest rung that is actually safe, not the most capable model overall.
• Latency & offline. No round-trip, no rate limit, no outage.

What this prototype is

This is an early prototype. The site, the search, and the task catalog are real, but no task has been evaluated yet — so there are deliberately no verdicts, model tables, recommended models, or “verified” dates anywhere. Publishing a verdict without a real eval run is exactly the mistake this product exists to avoid, so the catalog stays honest until the evidence exists.

The full design — task catalog, evidence runs, practitioner voting, the agent API/MCP — is laid out in How it works.

Start by browsing the catalog, or read how a verdict will read.