omm link
Re-verify and repair every installed model's links into every installed runner — or link the whole hub into a directory for an app omm doesn't support directly.
01 / 06
Overview
Reach for link after installing a new runner (so models installed before it get linked in too), or when scan's "aren't linked into an installed engine yet" note shows up. With no directory it always re-links every model into every installed runner — covering both never-linked and quietly-broken links, since it never trusts a cached linked flag. With a directory, it reuses the central file directly when possible instead of copying it, for an app that isn't one of the seven omm knows natively.
02 / 06
Options
Every flag this command accepts, and what it defaults to when you leave it out.
[directory]—Default: none — repairs known app linksAn extra directory to link every installed model's file into, for an app omm doesn't support directly.
--engineNAMEDefault: every engineOnly re-verify/repair links for this one engine.
--force—Default: offReclaim a destination omm doesn't recognize as its own by deleting and relinking it, instead of skipping it as a conflict.
03 / 06
Examples
From a plain search to something you'd put in a script.
Re-verify and repair every model's links into every installed runner.
$ omm linkOnly repair links into Ollama.
$ omm link --engine ollamaLink every installed model into a directory for an unsupported app.
$ omm link ~/my-runner/models04 / 06
A real run
Real omm link run, 2026-08-25: installed a real small model against a throwaway OMM_HOME, ran link for real (rewriting its actual Ollama symlink, exactly what link does), then uninstalled it immediately after capture — confirmed removed from the real ollama list too. "0 skipped (file missing)" is the message's real exact wording, not "missing" as an earlier version of this page guessed without running it.
06 / 06
If something goes wrong
Every message below is one this command actually prints. Find yours, read why it happened, then do the last line.
--engine must be one of: anythingllm, jan, koboldcpp, lmstudio, mstystudio, ollama, textgenwebui (got 'bogus').- why
- --engine only accepts a real engine key, and this run passed something else.
- what to do
- Use one of the seven listed in the message.
- source
- src/omm/cli.py:909-911
--engine only applies without a directory argument.- why
- --engine narrows which runner gets repaired; a directory argument means something different — linking into a location that isn't one of the seven known runners. The two can't be combined.
- what to do
- Use --engine alone to repair one runner's links, or a directory alone to link into a custom location.
- source
- src/omm/cli.py:6501-6502
Still stuck? Open an issue with the exact message you saw.
All commands
- omm searchFind a model across the curated catalog, HuggingFace and ModelScope.
- omm installDownload a model into the hub and link it into every installed runner.
- omm runChat with an installed model — in the terminal for Ollama, or by opening the app for GUI runners.
- omm recommendGet a model suggestion ranked for this machine's hardware, with an offer to install it.
- omm contributeBenchmark models in a loop, uploading telemetry to improve recommend for hardware like yours.
- omm setupRe-run the hardware scan and runner-install checklist, any time.
- omm scanPrint this machine's hardware, detected runners, and models — no flags needed.
- omm tuneGet recommended context length, GPU offload, threads, and batch size for a model.
- omm fitSee whether a model fits this machine's free memory right now, installed or not.
- omm helpShow omm's own command summary, or the full reference with --all.
- omm importAdopt .gguf files sitting in other apps' model directories into the omm hub.
- omm uninstallRemove a model and clean up its symlinks and manifests. Alias: rm.
- omm listShow every model omm has installed and which runners each is linked into. Alias: ls.
- omm infoShow full detail — repo, version, size, links, run commands — for one installed model.
- omm upgradeRefresh installed models against their source — only re-downloads what's actually changed. Alias: up.
- omm cleanupClean up leftover partial downloads and broken runner symlinks in one pass — no flags needed.
- omm verifyProve that an installed model actually loads and generates text on this machine.
- omm benchmarkLocal quality and speed smoke evidence for one or more installed models.
- omm updateReinstall omm from the latest source and refresh its recommendation data.
- omm settingView or change omm's settings — telemetry, outbound data, theme, update channel, and more.
- omm doctorDiagnose the omm install and Ollama links, read-only — no flags needed.
- omm engine installInstall one local AI runner program directly, skipping the setup checklist.
- omm logRead the local run log: what omm ran, when, and whether it worked.
- README — UsageEvery omm command, one line each.