Lmstudio Warm

opencode-lmstudio-warm

Deterministic LM Studio model pre-warm gate for opencode — loads and keeps the target model resident before every request, healing cold starts and mid-session TTL evictions.

Actively maintained

#opencode #opencode-plugin #lmstudio #lm-studio #llm #local-llm

GitHub repo npm page Homepage

GitHub stars

Monthly installs

828

301 in 7 days

Composite scoreSCORE

37.9

Multi-signal model

Last commit

7 hours ago

2026-08-20

Install and configure

opencode.json

Writes to this project's opencode.json — applies to this repository only.

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-lmstudio-warm@0.3.2"]
}

Writes to ~/.config/opencode/opencode.json — applies to every project.

~/.config/opencode/opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-lmstudio-warm@0.3.2"]
}

opencode loads npm dependencies through its embedded runtime on startup and caches them locally — no manual global install needed.

README.md

View original→

A bun-workspaces monorepo for deterministic LM Studio model pre-warming across coding-agent integrations. Every package warms a target model — makes sure it's already loaded and addressable in LM Studio — before it lets a completion request leave, instead of paying an in-flight cold-load latency spike or racing a "no model loaded" error.

Quick start: opt in with a config file, LM Studio starts cold, the first request warms the model before it leaves, and lms ps shows the model resident with no TTL

^{Scripted demo using the omp package — the status/log
lines are the plugins' real strings; the cold-load wait is shortened.}

Packages

Package	npm name	What it wires into
`packages/core`	`lm-studio-warm-core`	Runtime-agnostic core: cross-process lock, eviction planning, `lms` CLI client, model discovery, two-tier config loading. Not user-facing on its own.
`packages/omp`	`omp-lm-studio-warm`	`omp` extension — warms LM Studio models before every `lm-studio` completion stream.
`packages/pi`	`pi-lm-studio-warm`	`pi` extension — same gate, wired as a native `lm-studio` provider.
`packages/opencode`	`opencode-lmstudio-warm`	opencode plugin — gates every request via the `chat.params` hook; heals mid-session TTL evictions.

All four packages share the same WarmOptions config shape and the same cross-process lock at ~/.cache/lm-studio-warm/lock — see packages/core/README.md for the canonical configuration reference and lock semantics. Each wiring package's own README covers its install steps, config file location, and runtime-specific behavior.

Repo map

lm-studio-warm/
├── package.json                      # private workspace root; no publishable code
├── tsconfig.base.json · bunfig.toml
├── release-please-config.json · .release-please-manifest.json
├── .github/workflows/                # ci.yml (per-package matrix), release-please.yml, publish.yml
├── docs/assets/                      # demo assets — quickstart-<runtime>.{cast,gif}
└── packages/
    ├── core/              → npm lm-studio-warm-core
    ├── omp/               → npm omp-lm-studio-warm
    ├── pi/                → npm pi-lm-studio-warm
    └── opencode/          → npm opencode-lmstudio-warm

Development

bun install
bun run check      # typecheck + test, fanned out to every package
bun run typecheck
bun run test

CI (.github/workflows/ci.yml) runs a per-package check matrix plus a bun audit job.

See packages/*/README.md for package-specific documentation.

Releases

Versioning is automated with release-please in manifest mode, one component per package. Conventional commits drive per-package version bumps; no one edits a version number by hand.

Naming note: the opencode package keeps its historical npm spelling (opencode-lmstudio-warm, no hyphen in "lmstudio") — npm rejects new names that differ from an existing package only in punctuation, so the rename to match its siblings is not possible on the registry.

License

MIT. See LICENSE.