LocalAILocalAI
LocalAI 是「换 URL 就转本地」的多模态运行时:单二进制 OpenAI 兼容前置,底下接多引擎, 文本/语音/视觉/agent 负载同一栈。MIT、活跃维护(v4.8.0 官网直抓口径)。
LocalAI is the 'change the URL and go local' multimodal runtime: a single OpenAI-compatible binary fronting multiple engines, with text/voice/vision/agent workloads on one stack. MIT, actively maintained (v4.8.0 per the official site).
适合与不适合When to use it, when not to
适合:现有 OpenAI 客户端最小改动转本地、CPU-only 机器、要多模态负载同一运行时。 不适合:单一 LLM 极致吞吐(专用引擎更精);要机构级背书(个人主导)。
Fits: minimal-change local switch for existing OpenAI clients, CPU-only machines, multimodal workloads on one runtime. Not a fit: squeezing max throughput from a single LLM; needing institutional backing.
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 Mudler T.(意大利,个人主导开源项目)。Mudler T. (Italy, individual-led open-source project).
- 国内使用Direct access
- 大陆可直连direct OK localai.io 直连实测 200(2026-10-21)。localai.io 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);开源项目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 open-source project 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
多引擎(engines)架构接各家后端;官网口径「Run text, voice, vision, image, video, 3D and agent workloads through one open runtime」(直抓所见);模型/引擎矩阵未逐条清点。
A multi-engine architecture connecting various backends; official-site stance "Run text, voice, vision, image, video, 3D and agent workloads through one open runtime" (direct-fetched); the model/engine matrix not inventoried item by item.
- 运行位置Where it runs
单二进制 OpenAI 兼容 API 前置多后端;官网称亦兼容 Anthropic/Ollama/ElevenLabs API(直抓所见); GitHub 50,065★(API 核验 2026-10-21)。
A single binary: an OpenAI-compatible API fronting multiple backends; the site also claims Anthropic/Ollama/ElevenLabs API compatibility (direct-fetched); GitHub 50,065★ (API verified 2026-10-21).
- 本地文件Local files
官网口径「keep your data on your hardware」(直抓所见);全自托管。
Official-site stance "keep your data on your hardware" (direct-fetched); fully self-hosted.
- 关机后的任务Tasks after shutdown
常驻服务形态(单二进制 API server);CPU-only 模式官网点名(直抓所见),并发细节未逐条核验。
A resident-service form (single-binary API server); CPU-only mode is named on the official site (direct-fetched); concurrency details not verified item by item.
- 工具与扩展Tools & extensions
agent 负载在官方多模态叙事内;工具调用面与 MCP 支持未逐条核验。
Agent workloads sit inside the official multimodal narrative; the tool-calling surface and MCP support not verified item by item.
- 上下文与记忆Context & memory
上下文长度按后端引擎与部署配置变化;跨请求缓存与记忆线未逐条核验。
Context length varies by backend engine and deployment configuration; cross-request caching and memory lines not verified item by item.
- 权限与限制Permissions & limits
自托管无账号体系;API 鉴权由部署方自配;多引擎后端的凭据面(如接云 API)未逐条核验。
Self-hosted with no account system; API authentication configured by the deploying party; the credential surface of multi-engine backends (e.g. connecting cloud APIs) not verified item by item.
- 适合什么任务Best for
适合:一个二进制换 URL 就把现有 OpenAI 客户端转本地、要文本之外(语音/图像)多模态负载同一运行时。 不适合:只要单一 LLM 极致性能(专用引擎更精);要大厂官方背书(个人主导项目)。
Fits: switching existing OpenAI clients to local by just changing a URL with one binary; wanting text-plus (voice/image) multimodal workloads on one runtime. Does not fit: single-LLM maximum performance (dedicated engines are sharper); wanting big-institution backing (an individual-led project).
头号误解Top misconception
- 把「兼容多家 API」当「性能对齐各家官方」——兼容是接口面,性能跟后端引擎走Reading 'compatible with several APIs' as 'performance matches each vendor' — compatibility is the interface surface; performance follows the backend engine
- 个人主导项目的维护面——看仓库节奏与贡献者面(GitHub API 快照口径)Maintenance of an individual-led project — watch repo cadence and contributor breadth (GitHub API snapshot)
- 引擎矩阵复杂——具体模型跑不跑得动先查官方 engines 文档The engine matrix is complex — check the official engines docs before assuming a model runs
未知项清单Known unknowns
- engines 全矩阵与各引擎的模型支持面(未逐条清点)
- Anthropic/ElevenLabs 兼容面的完成度(官网口径直抓,细节未逐条核验)
- MCP 支持状态(未逐条核验)
- The full engines matrix and each engine's model-support surface (not inventoried item by item)
- The completeness of the Anthropic/ElevenLabs compatibility surfaces (official-site claims direct-fetched; details not verified item by item)
- MCP support status (not verified item by item)
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 | LocalAI · 官网(MIT/v4.8.0/多 API 兼容口径,2026-10-21 直抓核验)LocalAI · official site (MIT/v4.8.0/multi-API compatibility stance, direct-fetched and verified 2026-10-21) | https://localai.io |
| repo | mudler/LocalAI · 仓库(MIT;50,065★,最近推送 2026-10-29,GitHub API 核验)mudler/LocalAI · repo (MIT; 50,065★, last push 2026-10-29, GitHub API verified) | https://github.com/mudler/LocalAI |
| docs | LocalAI · 官方文档LocalAI · official docs | https://localai.io/docs |
| changelog | mudler/LocalAI · Releases(changelog)mudler/LocalAI · Releases (changelog) | https://github.com/mudler/LocalAI/releases |
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).
相关条目Related entries
- OllamaOllama同为 OpenAI 兼容本地栈:Ollama 模型库开箱顺滑,LocalAI 多模态与多 API 兼容更宽。Both OpenAI-compatible local stacks: Ollama's model library is smoother out of the box, LocalAI is broader on multimodal and multi-API compatibility.
- vLLMvLLM单一 LLM 服务化吞吐 vLLM 更精;LocalAI 的价值在「一个运行时多种负载」。vLLM is sharper on single-LLM service throughput; LocalAI's value is "one runtime, many workloads".
- OpenAI Agents SDKOpenAI Agents SDK另一层 · 框架与运行时Other layer · Frameworks & runtimes
- LlamaIndex FrameworkLlamaIndex Framework另一层 · 框架与运行时Other layer · Frameworks & runtimes
本页由This page is generated from the ai-handbook 数据层生成(CC BY 4.0)。data layer (CC BY 4.0).
方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.