Runnability, not guesswork
An AI Hardware Score (0–100) names your bottleneck, and every model gets a yes / slow / no verdict with an estimated tok/s range. Unknown hardware says “unknown” — never a made-up number.
Know which local LLMs will run on your machine
local-llmup scores your hardware, then gives every model a straight answer — yes, slow, or no, with an estimated tokens-per-second range — before you download a thing. Then it pulls, serves, and migrates memory, all locally.
npm install -g local-llmup
10 commands - yes/slow/no verdicts - est. tok/s - local-first workflow
One-command lifecycle
`up`, `down`, `switch`, and `doctor` coordinate the local server lifecycle through Ollama with safer defaults and consistent diagnostics.
Portable memory switching
Migrate local memory from one model to another when you change quality or latency targets, while keeping conversational continuity.
Get running in a few commands
Use the defaults for quick setup, then switch to a specific model when ready. These are examples from a typical local workflow.
Install and rank
npx local-llmup
Prints ranked models with a verdict and est. tok/s, plus a top-pick command.
Will it run?
local-llmup can-run llama3.1:8b
Answers yes / slow / no with est. tok/s. Exits non-zero only for no, so it scripts.
Pull and serve
local-llmup up qwen3:30b-a3b
Pulls the model through Ollama and brings up the local server endpoint.
Chat
local-llmup chat -m qwen3:30b-a3b
Starts an interactive prompt against your active or selected model.
Diagnostics
local-llmup doctor
Scores your hardware, names the primary bottleneck, and reports actionable issues.