← Runners

LM StudioLM Studio

Element Labs

闭源免费Closed source, free 桌面应用Desktop application GGUFGGUF MLXMLX llama.cppllama.cpp 本地agentLocal agent partial lifecycle: active 核验Checked 2026-10-30

English summaries and interface labels are available. Detailed source notes below are currently in Chinese.

LM Studio 是桌面跑本地模型的主流图形入口：应用内下载模型、双运行时（MLX + llama.cpp）、内置本地服务器。 2026-10 快照：官网主推 Bionic——本地 agent 叙事已盖过纯聊天。诚实标注：应用闭源；条款页动态加载未能直抓全文。

LM Studio is the mainstream graphical entry for local models on the desktop: in-app model downloads, dual runtimes (MLX + llama.cpp), built-in local server. 2026-10 snapshot: the site now leads with Bionic — a local-agent narrative that overshadows plain chat. Honest notes: the app is closed-source; the terms page loads dynamically and could not be direct-fetched.

适合与不适合When to use it, when not to

适合：不想碰命令行的本地模型用户、要在同一应用试 agent 与语音转写。不适合：服务器无头部署；对许可条款有硬合规要求（本站未能直抓全文，只挂官网口径）。

Fits: local-model users who dislike the CLI, trying agents and voice transcription in one app. Not a fit: headless server deployment; hard compliance needs on license terms (full text could not be direct-fetched).

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-21

出品属地Origin: 国外出品foreign-made　Element Labs（美国）。Element Labs (US).
国内使用Direct access: 大陆可直连direct OK　lmstudio.ai 直连实测 200（2026-10-21）。lmstudio.ai measured 200 direct (2026-10-21).
备案状态CAC filing: 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-21 检索）；该产品未面向大陆公众提供服务，备案不适用。not in the CAC public registry (searched 2026-10-21); the product does not serve the mainland public, so filing does not apply.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness: 应用内直接下载最新本地 LLM（官网直抓所见）；GGUF 为主，MLX 与 llama.cpp 双运行时（官网原文「Powered by the LM Studio runtime, with MLX and llama.cpp under the hood」）。
Download the latest local LLMs directly in-app (as seen on the official site, direct-fetched); GGUF-first, with MLX and llama.cpp dual runtimes (official wording: "Powered by the LM Studio runtime, with MLX and llama.cpp under the hood").
运行位置Where it runs: 桌面应用（macOS/Windows/Linux）；官网现主推「Bionic」agent（本地 agent for work and code）—— 产品叙事已从纯聊天扩到 agent，建档快照口径。
Desktop app (macOS/Windows/Linux); the site now leads with the "Bionic" agent (a local agent for work and code) — the product narrative has expanded from pure chat to agent, per the filing snapshot.
本地文件Local files: 模型与对话落本机；官网明写语音转写「processed locally and never leaves your device」（直抓所见）。
Models and conversations land on your machine; the site states plainly that speech transcription is "processed locally and never leaves your device" (direct-fetched).
关机后的任务Tasks after shutdown: 内置本地服务器（OpenAI 兼容端点）可常开供外部客户端用；默认开关与鉴权细节未逐条核验。
A built-in local server (OpenAI-compatible endpoint) can stay on for external clients; default switch and authentication details not verified item by item.
工具与扩展Tools & extensions: Bionic agent 线含文档编辑/自动化/计算机控制（官网直抓所见）；MCP 支持状态未逐条核验。
The Bionic agent line includes document editing/automation/computer control (official site, direct-fetched); MCP support status not verified item by item.
上下文与记忆Context & memory: 聊天与 agent 会话的上下文管理在应用内；跨会话记忆与长期记忆未见官方声明。
Context management for chat and agent sessions is in-app; cross-session memory and long-term memory — no official statement seen.
权限与限制Permissions & limits: 桌面应用权限随系统；agent 的计算机控制面权限边界未逐条核验。
Desktop-app permissions follow the system; the permission boundary of the agent's computer-control surface not verified item by item.
适合什么任务Best for: 适合：要图形界面跑本地模型、想在同一应用里试 agent（Bionic）。不适合：要无头服务器部署（用 llama.cpp/vLLM）；要读全许可条款（条款页动态加载，本站未能直抓）。
Fits: running local models through a graphical interface, trying the agent (Bionic) in the same app. Does not fit: headless server deployment (use llama.cpp/vLLM); reading the full license terms (the terms page loads dynamically and this site could not direct-fetch it).

头号误解Top misconception

把「免费」读成「开源」——应用闭源，开源的是底层运行时Reading 'free' as 'open source' — the app is closed; the open parts are the underlying runtimes
只记得它是聊天壳——官网现主推 Bionic agent，产品面已扩Remembering it only as a chat shell — the site now leads with the Bionic agent; the product surface has expanded
把「语音不出设备」推广到全部功能——官网只对语音转写做了这个承诺Generalizing 'voice never leaves the device' to all features — the site makes that promise only for voice transcription

未知项清单Known unknowns

条款页全文（动态加载，直抓为空——不猜内容）
Bionic agent 的计算机控制权限模型（未逐条核验）
跨会话记忆能力（未见官方声明）

The full terms page (loads dynamically, direct fetch came back empty — content not guessed)
The computer-control permission model of the Bionic agent (not verified item by item)
Cross-session memory capability (no official statement seen)

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-30，内容更新日content updated on 2026-10-21。

类型Type	名称Name	链接Link
docs	LM Studio · 官网（Bionic 叙事与本地承诺，2026-10-21 直抓核验）LM Studio · official site (the Bionic narrative and local promises, direct-fetched and verified 2026-10-21)	https://lmstudio.ai
docs	LM Studio · 官方文档LM Studio · official docs	https://lmstudio.ai/docs
changelog	LM Studio · 官方博客LM Studio · official blog	https://lmstudio.ai/blog
repo	LM Studio · GitHub 组织（llama.cpp/MLX 集成侧仓库）LM Studio · GitHub org (the llama.cpp/MLX integration-side repos)	https://github.com/lmstudio-ai

实测记录Measured runs

本档未跑吞吐与显存实测；结论只覆盖运行时能力与许可。

This site has not completed measured runs. The closest test protocol is the code-side protocols/framework/_protocol.md (the local-endpoint and credential-boundary sections are reusable; no runner-specific protocol exists on this site yet).

第一次怎么用

1. 安装桌面应用，先在应用内搜索并下载模型；注意应用闭源。 2. 选择 MLX 或 llama.cpp 运行时：Apple 硅优先 MLX，其他平台用 llama.cpp。 3. 启动内置本地服务器，把客户端指到它的 OpenAI 兼容端点。 4. 条款页动态加载，未能直抓全文；商用前在官网读完条款并存档。 5. 显存不足时先降量化档，再调上下文长度。

本地服务器默认端口以应用界面为准；
客户端 base_url 指向本机端点，模型名填应用里显示的模型 ID。