local-models
/local-models
Przypomina zasady pracy z lokalnymi modelami LM Studio przez hub: wysyłanie żądań, ładowanie i wymianę modeli.
- Autor
- haedolabs ↗
- Repozytorium
- local-models-hub ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 12.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 2
Brak sygnałów. Repozytorium zmienione 12.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Przypomina zasady pracy z lokalnymi modelami LM Studio przez hub: wysyłanie żądań, ładowanie i wymianę modeli.
Oryginalny opis (EN)
The rules for using the local models (LM Studio) on this Mac through the local-models hub. Use whenever a task involves a local model, LM Studio, port 1234 or 1235, a local agent or batch over a project's files, loading or unloading a model, or the question of which model is loaded. Covers where to send requests, the MCP tools (status, guide, ask, batch, job_status, swap_check), leases, swaps, and what happens when a model crashes.
Jak działa
SKILL.md bez nagłówków drugiego poziomu · 14 linii, 496 słów
Podgląd SKILL.md
--- name: local-models description: The rules for using the local models (LM Studio) on this Mac through the local-models hub. Use whenever a task involves a local model, LM Studio, port 1234 or 1235, a local agent or batch over a project's files, loading or unloading a model, or the question of which model is loaded. Covers where to send requests, the MCP tools (status, guide, ask, batch, job_status, swap_check), leases, swaps, and what happens when a model crashes. --- # Local models on this Mac, through the hub - **Send requests to `http://localhost:1235/v1`** (the hub's gateway: LM Studio's own API, same paths), never directly to LM Studio's port 1234. The gateway runs admission control across all clients, attributes each request to its project and session, and logs metadata only, never content. Optional headers: `X-Project` (a name), `X-Kind` (`batch`, `chat`, `ask`, `eval`, `test`), `X-Step` (what this request is for, shown on the dashboard). **A batch script also sends `X-Run` (the run's name, the same every time it runs), `X-Run-Total` (units in the whole run) and `X-Run-Done` (units completed)**: the dashboard then shows a progress bar and an ETA, measured on this Mac (`guide section=progress`). - **Names, three levels:** your project's name on the hub's screens is set once with the `name` tool (`name="research"`, `formerly=[...]` for an `X-Project` your scripts used to send); a batch's name is the `X-Run` header; a job's is the phase in `X-Step`. "Call this project X" means the `name` tool, not `X-Run`. - **Check before you assume.** The `status` tool reports the loaded models, their total concurrent slots, the clients using them now, and any model down after a crash. The `guide` tool gives the full rules (`section=...` for one part). - **Bulk reading jobs go to local agents**: the `ask` tool (one question, read-only tools scoped to this project's folder, blocks up to two minutes) and `batch` (many jobs in parallel; poll with `job_status`). Their answers are leads to verify, never verdicts: models fabricate quotes and line numbers. `think=false` makes lookups and extraction several times faster; keep thinking on for checks and judgment. - **A batch should keep as many requests in flight as the loaded replicas have slots** (the `status` tool reports it; the gateway queues the rest). A long run that pauses between requests holds a lease (`guide section=lease`). - **Never stop LM Studio's server. Never unload or reload a model without `swap_check`**; if it names anyone, do not swap. The simplest swap is `localmodels use <model>` in a terminal. - **If a model crashes, the instance manager reloads it** within a minute and the gateway holds new requests for it meanwhile; a request that was in flight is resent once, and if the model is still down the gateway returns 503 with `Retry-After`. The instance manager also recycles each instance before the token count at which the engine has crashed, one instance at a time, so a long batch keeps running. - **Maintenance:** when the hub announces it (a line in the session, the `X-Hub-Maintenance` header, or the `status` tool), do not start a local-model batch that would run past its start; a `503` with `Retry-After` during it means sleep that long and resume (`guide section=maintenance`). - **See it:** the dashboard at http://localhost:1235 (`localmodels open`), or `localmodels` in a terminal. - **If the gateway is down** (`status` says so): `localmodels start`. If the tools are missing entirely, the hub is not set up on this Mac yet: `/local-models:setup`.
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · haedolabs/local-models-hub | 96 | MIT |
| glebis/claude-skills | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki a3487429a496 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100local-model-orchestrator
Skanuje, objaśnia, dobiera, instaluje i wywołuje lokalne modele AI w Ollama, cache Hugging Face, MLX Whisper, whisper.cpp i LM Studio, z naciskiem na modele zoptymalizowane pod Apple Silicon. - 100ai-model-router
Automatycznie przełącza między dwoma skonfigurowanymi modelami: lokalnym do prostych zadań i chmurowym do złożonych, wykrywa Ollama i LM Studio oraz wymusza model lokalny przy danych wrażliwych. - 100agent-platform-model-registry
Zarządza modelami w Agent Platform Model Registry: wysyłanie, listowanie, opis, aktualizacja i usuwanie modeli uczenia maszynowego oraz ich wersji. - 100agent-platform-model-registry
Zarządza modelami w Agent Platform Model Registry: wysyłanie, listowanie, opis, aktualizacja i usuwanie modeli uczenia maszynowego oraz ich wersji. - 100docker-model
Uruchamia lokalne modele AI przez Docker Model Runner (CLI docker model): pobieranie i uruchamianie modeli, endpointy zgodne z OpenAI i integracja z Compose przez element models. - 100local-model-onboarding
Wprowadza nowy model lokalny do stanowiska ewaluacyjnego na Mac Studio: ocena wykonalności sprzętowej, wstępna ocena jakości przez OpenRouter, ollama pull, rejestracja w models.yml i włączenie do ciągłej rotacji.