← Runners

OllamaOllama

Ollama Inc.

开源Open source MITMIT 本地推理Local inference GGUFGGUF OpenAI兼容OpenAI-compatible 模型库Model library partial lifecycle: active 核验Checked 2026-10-30

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

Ollama 是「本地跑开源模型」的事实标准入口：一条命令装完，从官方模型库一键拉模型，本机起一个 OpenAI 兼容 API。诚实标注：工具 MIT，但权重许可跟着模型走；官方 README 明写还有云端线。

Ollama is the de-facto entry point for running open models locally: one command installs it, the official model library pulls models in one click, and an OpenAI-compatible API comes up on your machine. Honest notes: the tool is MIT but weight licenses follow each model; the official README explicitly mentions a cloud line.

适合与不适合When to use it, when not to

适合：快速把 DeepSeek/Qwen/Llama 这类开源模型跑起来、给本地应用接 OpenAI 兼容端点、桌面与服务器同一套命令。不适合：高并发服务化调度（vLLM/SGLang 更合适）；对云端线数据边界有硬要求但没逐条核验过。

Fits: getting DeepSeek/Qwen/Llama running with one command, OpenAI-compatible local endpoints, desktop and server alike. Not a fit: high-concurrency serving (vLLM/SGLang), hard data-boundary requirements on the cloud line.

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　Ollama Inc.（美国，YC 系）。Ollama Inc. (US, YC-backed).
国内使用Direct access: 大陆可直连direct OK　ollama.com 与 docs.ollama.com 直连实测 200（2026-10-21）。ollama.com and docs.ollama.com 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: 官方模型库（ollama.com/search 直抓 200）一键拉取开源模型；GGUF 与 safetensors 可导入（docs.ollama.com/import 直抓核验）；模型库目录随上架变化，未逐条清点。
One-click pulls of open-source models from the official model library (ollama.com/search direct-fetched 200); GGUF and safetensors importable (docs.ollama.com/import direct-fetch verified); the library catalog shifts with listings, not inventoried item by item.
运行位置Where it runs: 本地进程（CLI + API server，默认 11434 端口）；官方 README 声明「locally or in the cloud」—— 云端线（Ollama Cloud）存在，细节未逐条核验。GitHub 185,274★（API 核验 2026-10-21）。
A local process (CLI + API server, port 11434 by default); the official README states "locally or in the cloud" — a cloud line (Ollama Cloud) exists, details not verified item by item. GitHub 185,274★ (API verified 2026-10-21).
本地文件Local files: 模型权重与对话数据默认落本机；云端线（Ollama Cloud）的存储与处理边界未逐条核验。
Model weights and conversation data land on your machine by default; the storage and processing boundaries of the cloud line (Ollama Cloud) not verified item by item.
关机后的任务Tasks after shutdown: API server 常驻即承载后台推理；官方文档有 keep-alive/并发说明页，细节未逐条核验。
The resident API server carries background inference; official docs have keep-alive/concurrency pages, details not verified item by item.
工具与扩展Tools & extensions: 官方文档含 tool calling 页（docs.ollama.com 导航直抓所见）；MCP 支持状态未核验到官方声明。
Official docs include a tool-calling page (as seen in the docs.ollama.com navigation, direct-fetched); MCP support status — no official statement verified.
上下文与记忆Context & memory: 上下文长度按 Modelfile/参数配置（Modelfile Reference 在官方文档导航内）；跨会话记忆未见官方声明。
Context length is set by Modelfile/parameters (the Modelfile Reference is in the official docs navigation); cross-session memory — no official statement seen.
权限与限制Permissions & limits: 本机运行无账号即可用（API server 默认本机监听）；云端线需登录，其权限面未逐条核验。
Runs locally with no account needed (the API server listens locally by default); the cloud line requires login, its permission surface not verified item by item.
适合什么任务Best for: 适合：一条命令把开源模型跑起来、要 OpenAI 兼容本地 API、桌面与服务器通吃。不适合：要极致吞吐/服务化调度（那是 vLLM/SGLang 的活）；要逐条核验云端线数据边界。
Fits: getting an open-source model running with one command, wanting an OpenAI-compatible local API, desktop and server alike. Does not fit: maximum throughput/service scheduling (that's vLLM/SGLang's job); item-by-item verification of the cloud line's data boundaries.

头号误解Top misconception

把「MIT 的 Ollama」当「模型免费商用」——权重许可跟着模型走，不跟工具走Reading 'MIT Ollama' as 'models free for commercial use' — weight licenses follow the model, not the tool
只看到本地线——官方 README 明写云端线，数据边界要单独查Seeing only the local line — the official README explicitly states a cloud line; its data boundary needs a separate check
用默认上下文跑长文——上下文按 Modelfile/参数配置，默认值随模型变Running long documents on default context — context follows the Modelfile/params and varies by model

未知项清单Known unknowns

Ollama Cloud 的数据存储与处理边界（未逐条核验）
MCP 支持状态（未核验到官方声明）
模型库完整目录与各家权重的商用条款（未逐条清点）

Ollama Cloud's data storage and processing boundaries (not verified item by item)
MCP support status (no official statement verified)
The full model-library catalog and the commercial terms of each weight (not inventoried 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	ollama/ollama · LICENSE（MIT，2026-10-21 直抓核验）ollama/ollama · LICENSE (MIT, direct-fetched and verified 2026-10-21)	https://github.com/ollama/ollama/blob/main/LICENSE
repo	ollama/ollama · 仓库（185,274★，最近推送 2026-10-30，GitHub API 核验）ollama/ollama · repo (185,274★, last push 2026-10-30, GitHub API verified)	https://github.com/ollama/ollama
docs	Ollama · 官方文档（import/Modelfile/tool calling 页在导航内）Ollama · official docs (import/Modelfile/tool-calling pages in the navigation)	https://docs.ollama.com
changelog	ollama/ollama · Releases（changelog）ollama/ollama · Releases (changelog)	https://github.com/ollama/ollama/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。 models 站的 GGUF 索引口径（llama.cpp 格式源头）与本档案 model_access 维衔接。

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). The models site's GGUF index framing (llama.cpp as the format source) connects with this dossier's model_access axis.

第一次怎么用

1. 安装后先跑 ollama --version，确认服务已启动。 2. 从官方模型库拉一个模型；权重许可跟着模型走，商用前单独查。 3. 本机默认起 OpenAI 兼容端点，客户端把 base URL 指过来即可。 4. 显存不够时先降量化档，再考虑 CPU 卸载；不要直接换卡。 5. 官方 README 另写有云端线；本地与云端是两条计费与数据路径。

ollama pull <model>
ollama run <model>
# OpenAI 兼容端点默认在本机；客户端 base_url 指向它