← Coding

ContinueContinue

Continue

编程Coding 开源Open source 多模型Multiple models VSCodeVSCode partial lifecycle: archived 核验Checked 2026-10-23

开源、模型完全自配的 IDE 扩展。但它已停更：Continue 被 Cursor 收购，上游仓库转入 read-only，官网只剩收购公告 —— 代码仍可自由使用，后续不会再有更新。

An open-source IDE extension with fully self-configured models — VS Code extension, JetBrains plugin and a cn CLI over one agent core, with models configured freely in config.yaml. The official model-provider list spans Anthropic, Azure AI Foundry, Amazon Bedrock, DeepSeek, Gemini, Mistral, Ollama (local models), OpenAI, Vertex AI, xAI and "More (30 items)".

The extension itself is free; the commercial line is a separate product. The repository README states "Apache 2.0 © 2023-2026 Continue Dev, Inc." and there is no license fee for the extension — model calls are yours to pay (BYOK, or local Ollama). The commercial line is now Continuous AI at hub.continue.dev/pricing: Starter $3 / million tokens pay-as-you-go, Team $20 / seat / month (including $10 in credits per seat), Company custom (SAML/OIDC, BYOK, SLA).

But the official status has changed. Continue was acquired by Cursor; the homepage continue.dev now serves an acquisition notice, the old Hub pricing page continue.dev/pricing returns 404, and the repository is marked read-only. The last releases are the final 2.0.0 line for the VS Code extension, CLI and JetBrains plugin.

适合与不适合When to use it, when not to

需要模型自由（多 provider，含 Ollama 本地模型）而不想被单一厂商锁定的人。
已在 VS Code / JetBrains 里工作、希望用开源可审计扩展的人。
需要把同一 agent 用到终端与 CI（cn -p）的人。

不适合需要持续维护与长期演进的场景 —— 上游仓库已 read-only，公司被 Cursor 收购，后续只会以「代码仍可自由使用」的形态存在。也不适合依赖官方支持与持续修复的人。

A fit for three groups:

People who need model freedom (many providers, including local models via Ollama) and do not want to be locked to one vendor.
People already working in VS Code / JetBrains who want an open-source, auditable extension.
People who want the same agent in the terminal and in CI (cn -p).

Not a fit for scenarios needing ongoing maintenance and long-term evolution — the upstream repository is read-only and the company was acquired by Cursor, so it persists only in a "code is still freely usable" form. Also not a fit for anyone relying on official support and continued fixes.

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 2026-10-20

出品属地Origin: 国外出品foreign-made　国外主体（网信办公示名单未见对应备案主体）。foreign entity (no matching filing entity in the CAC registry).
国内使用Direct access: 大陆可直连direct OK　官网直连实测 200。official site measured 200 direct.
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（2026-10-20 检索 13 批附件，https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm）；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 (13 batch attachments searched 2026-10-20, https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm); 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: 模型完全自配，这是它最核心的定位。

官方 model providers 页列出的 provider：

Anthropic
Azure AI Foundry
Amazon Bedrock
DeepSeek
Gemini
Mistral
Ollama（本地模型）
OpenAI
Vertex AI
xAI

该页还指向「More（30 items）」。

配置写在 config.yaml 的 models，字段为 provider + model + apiBase。

可设 roles（chat / autocomplete / embed / rerank / edit / apply / summarize），以及 capabilities（tool_use 为 Agent 模式必需，另有 image_input）。

与 ./cursor.md 对照：Cursor 只能用其目录内模型，Continue 可指任意 provider。

各家 provider 换用后的 tool calling 质量，官方页面未给出对比。
Models are fully self-configured — this is its most core positioning.

Providers listed on the official model providers page:

Anthropic
Azure AI Foundry
Amazon Bedrock
DeepSeek
Gemini
Mistral
Ollama (local models)
OpenAI
Vertex AI
xAI

That page also points to "More (30 items)".

Configuration lives in config.yaml under models, with fields provider + model + apiBase.

roles (chat / autocomplete / embed / rerank / edit / apply / summarize) can be set, plus capabilities (tool_use is required for Agent mode; image_input also exists).

Contrast with ./cursor.md: Cursor can only use models in its catalog; Continue can point at any provider.

Tool-calling quality after switching across providers: the official pages give no comparison.
运行位置Where it runs: IDE 扩展形态，不是独立编辑器。

官方列出的形态有 VS Code 扩展、JetBrains 插件，另有 CLI（cn）。

官方文档首句："Open source AI code assistant for VS Code and JetBrains"

仓库 README 原文建议："We recommend using the Continue CLI instead of the JetBrains plugin"

三条形态共享同一 agent 核心。官方 CLI 页原文："the same agent that powers the Continue IDE extensions"。CLI 需 Node.js 20+ 或自带运行时。

本条目只写扩展形态，终端形态的能力见 tools 与 background。
An IDE-extension form, not a standalone editor.

The officially listed forms are the VS Code extension and the JetBrains plugin, plus a CLI (cn).

Official docs first sentence: "Open source AI code assistant for VS Code and JetBrains"

Official README recommendation: "We recommend using the Continue CLI instead of the JetBrains plugin"

The three forms share the same agent core. Official CLI page wording: "the same agent that powers the Continue IDE extensions". The CLI needs Node.js 20+ or a bundled runtime.

This entry covers only the extension form; the terminal form's capabilities are in tools and background.
本地文件Local files: 在 IDE 工作区内读写，改动以 diff 形式呈现。

上下文靠显式引用，不靠隐式全量索引。 官方 Chat 上下文页列出 @Files、@Terminal、@Git Diff 等 context providers，以及高亮代码（VS Code cmd/ctrl+L、JetBrains cmd/ctrl+J）与当前活动文件。

两个配置目录：MCP 为 .continue/mcpServers/，规则为 .continue/rules。

代码库语义索引的实现方式官方未说明 （已查 docs.continue.dev 首页与 Chat 上下文选择两页）。
Reads and writes inside the IDE workspace; changes are presented as diffs.

Context comes from explicit references, not implicit full indexing. The official Chat context page lists context providers such as @Files, @Terminal, @Git Diff, plus highlighted code (VS Code cmd/ctrl+L, JetBrains cmd/ctrl+J) and the current active file.

Two config directories: MCP at .continue/mcpServers/, rules at .continue/rules.

How the codebase semantic index is implemented is not stated by the official side (checked the docs.continue.dev homepage and the Chat context-selection pages).
关机后的任务Tasks after shutdown: 扩展形态无「关机续跑」能力。

它就是编辑器 / IDE 插件，随 IDE 进程存活，IDE 关掉即结束。

官方文档也未描述云端执行面（核验：已查 docs.continue.dev 首页、 Agent quick-start、CLI quickstart 三页）

唯一异步形态是 CLI 的 headless。 cn -p "..." 单次跑完输出到 stdout。

官方原文："Use this in scripts, CI/CD, and git hooks"
The extension form has no "keeps running after shutdown" capability.

It is an editor / IDE plugin, alive with the IDE process; close the IDE and it ends.

The official docs also describe no cloud execution surface (verified: checked the docs.continue.dev homepage, Agent quick-start and CLI quickstart pages)

The only async form is the CLI's headless mode. cn -p "..." runs once and outputs to stdout.

Official wording: "Use this in scripts, CI/CD, and git hooks"
工具与扩展Tools & extensions: 工具面是「三档模式 + MCP」。

官方 Agent 页写模式切换下有三个选项：

Chat（无工具）
Plan（只读工具）
Agent（全部工具）

Cmd/Ctrl + . 循环切换。

MCP 官方支持（官方 MCP 深潜页）：在 config 的 mcpServers 配置，可放 .continue/mcpServers/ 目录，传输支持三种：

stdio
sse
streamable-http

可直接复用 Claude / Cursor / Cline 的 JSON MCP 配置。

官方强调："MCP can only be used in the agent mode"

CLI 侧另有 --mcp <slug>、--agent <slug>、--rule。扩展机制还有 context providers、rules 与自定义 prompt。
The tool surface is "three mode tiers + MCP".

The official Agent page shows three options under mode switching:

Chat (no tools)
Plan (read-only tools)
Agent (all tools)

Cmd/Ctrl + . cycles through them.

MCP is officially supported (official MCP deep-dive page): configured under mcpServers in config, can live in the .continue/mcpServers/ directory; three transports supported:

stdio
sse
streamable-http

JSON MCP configs from Claude / Cursor / Cline can be reused directly.

The official stresses: "MCP can only be used in the agent mode"

The CLI side additionally has --mcp <slug>, --agent <slug>, --rule. Extension mechanisms also include context providers, rules and custom prompts.
上下文与记忆Context & memory: 上下文是显式选择式，不是自动索引式（见 local_files）。

会话层面：CLI 支持 --resume 重放，官方 CLI headless 页写 cn -p --resume 会

官方原文："replays the previous session's history"

跨会话长期记忆由 rules、context providers 与 assistant 配置承载； config.yaml 有 data 字段。

上下文窗口大小跟随所选模型，官方在已查页面未给出统一数字（已查 Agent quick-start 与 config.yaml 参考两页）。
Context is explicitly selected, not auto-indexed (see local_files).

At the session level: the CLI supports --resume replay; the official CLI headless page says cn -p --resume

official wording: "replays the previous session's history"

Cross-session long-term memory is carried by rules, context providers and assistant configuration; config.yaml has a data field.

Context-window size follows the chosen model; the official gives no unified number on the pages checked (checked the Agent quick-start and config.yaml reference pages).
权限与限制Permissions & limits: 默认逐工具询问，可策略化放开。

IDE 侧，官方 Agent quick-start 原文："By default, Agent mode will ask permission when it wants to use a tool"

tool policies 可把特定工具设为自动或排除。 IDE 里可点 Continue 放行，或点 Cancel 拒绝。

CLI 侧口径不同。官方 headless 页原文："tools that would normally prompt for approval (ask permission) are automatically excluded"；要写文件必须显式 --allow Write --allow Edit（或 --allow "*"），另有 --readonly（Plan）、--auto、--exclude。

telemetry：仓库 README 写最终 2.0.0 版「removing anonymous telemetry」。

数据是否用于训练，官方已查页面未说明。
Asks per tool by default; policies can loosen it.

On the IDE side, official Agent quick-start wording: "By default, Agent mode will ask permission when it wants to use a tool"

tool policies can set specific tools to automatic or excluded. In the IDE, click Continue to allow or Cancel to deny.

The CLI side differs. Official headless page wording: "tools that would normally prompt for approval (ask permission) are automatically excluded"; to write files you must explicitly pass --allow Write --allow Edit (or --allow "*"), plus --readonly (Plan), --auto, --exclude.

telemetry: the repository README says the final 2.0.0 release came "removing anonymous telemetry".

Whether data is used for training: not stated on the official pages checked.
适合什么任务Best for: 适合三类人：

需要模型自由（多 provider，含 Ollama 本地模型）而不想被单一厂商锁定的人。
已在 VS Code / JetBrains 里工作、希望用开源可审计扩展的人。
需要把同一 agent 用到终端与 CI（cn -p）的人。

不适合需要持续维护与长期演进的场景 —— 上游仓库已 read-only，公司被 Cursor 收购，后续只以「代码仍可自由使用」的形态存在。

也不适合依赖官方支持与持续修复的人。
Fits three kinds of people:

those needing model freedom (multi-provider, including Ollama local models) without single-vendor lock-in.
those already working in VS Code / JetBrains who want an open-source auditable extension.
those needing the same agent in the terminal and CI (cn -p).

Not a fit for scenarios needing continuous maintenance and long-term evolution — the upstream repository is read-only, the company was acquired by Cursor, and it will only persist as "the code remains freely usable".

Also not a fit for people depending on official support and continuous fixes.

头号误解Top misconception

以为 Continue 仍在持续维护 —— 仓库 README 已标 no longer actively maintained and is read-onlyAssuming Continue is still actively maintained — the repo README now states it is "no longer actively maintained and is read-only for all users".
以为 Hub 商业线旧价目仍有效 —— 官网已被收购公告替换，原定价页 continue.dev/pricing 已 404Assuming the old Hub commercial price list still applies — the site is now an acquisition notice and continue.dev/pricing returns 404.
以为模型自带额度 —— 扩展本身免费开源，模型费要自备 provider 或本地 OllamaAssuming models come with included quota — the extension itself is free and open source, but model costs are yours (bring your own provider or run local Ollama).

未知项清单Known unknowns

以下五项都是官方未给出的，不是没查：

仓库归档的具体日期与收购完成时间 —— 官方未给出确切日期（已查仓库 README、官网公告与 Releases）
原 Continue 账户 / 订阅的去向与数据处理 —— 官网 FAQ 仅列出问题，答案官方未公开
代码库语义索引的实现方式 —— 官方未说明（已查 docs.continue.dev 首页与 Chat 上下文选择两页）
数据是否用于模型训练 —— 官方已查页面未说明（已查仓库 README 与文档站首页）
各 provider 换用后 tool calling 质量的官方对比

All five items below are things the official does not give — not things left unchecked:

The repository's exact archival date and acquisition completion time — the official gives no exact dates (checked repository README, the official-site notice and Releases)
The fate of old Continue accounts / subscriptions and data handling — the official-site FAQ only lists the questions; answers are not published
How the codebase semantic index is implemented — not stated by the official (checked the docs.continue.dev homepage and the Chat context-selection page)
Whether data is used for model training — not stated on the official pages checked (checked repository README and the docs-site homepage)
Any official comparison of tool-calling quality across providers

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 2026-10-23，内容更新日content updated on 2026-10-08。

类型Type	名称Name	链接Link
repo	Continue · 仓库 README（三形态、Apache-2.0、read-only 公告）Continue · repository README (three forms, Apache-2.0, read-only notice)	https://github.com/continuedev/continue
changelog	Continue · Releases（最终 2.0.0 与 v2.1.0-vscode，2026-10-08 取到）Continue · Releases (final 2.0.0 and v2.1.0-vscode, fetched 2026-10-08)	https://github.com/continuedev/continue/releases
docs	Continue · 官网（被 Cursor 收购公告）Continue · official site (acquisition-by-Cursor notice)	https://continue.dev
docs	Continue · 文档站首页（开源、VS Code / JetBrains、CLI cn）Continue · docs-site homepage (open source, VS Code / JetBrains, CLI cn)	https://docs.continue.dev
docs	Continue Docs · MCP 深潜（三种传输、仅 agent 模式）Continue Docs · MCP deep dive (three transports, agent mode only)	https://docs.continue.dev/customize/deep-dives/mcp
docs	Continue Docs · Model providers（含 Ollama 本地模型与 30+ provider）Continue Docs · Model providers (incl. Ollama local models and 30+ providers)	https://docs.continue.dev/customize/model-providers
docs	Continue Docs · Agent quick-start（默认逐工具询问权限）Continue Docs · Agent quick-start (asks per-tool permission by default)	https://docs.continue.dev/ide-extensions/agent/quick-start
docs	Continue Docs · CLI quickstart（cn 安装、headless、--auto / --readonly）Continue Docs · CLI quickstart (cn install, headless, --auto / --readonly)	https://docs.continue.dev/cli/quickstart
docs	Continue Docs · CLI headless（ask 权限自动排除、--allow）Continue Docs · CLI headless (ask permission auto-excluded, --allow)	https://docs.continue.dev/cli/headless-mode
pricing	Continue · 商业线定价页 hub.continue.dev/pricing（现 Continuous AI）Continue · commercial-line pricing page hub.continue.dev/pricing (now Continuous AI)	https://hub.continue.dev/pricing

实测记录Measured runs

本档未跑端到端编码任务；能力结论来自官方文档，编码向测量协议见 protocols/coding/_protocol.md。

实测建议：重点测两点。

同一 config.yaml 在 IDE 扩展与 CLI（cn）两形态下的行为一致性
headless 模式对 ask 权限工具自动排除的边界 —— 官方明确「没人批准就自动排除」，这是最容易被脚本化用户踩到的一点。

This site has not completed measured runs. The test protocol is at protocols/coding/_protocol.md.

Suggested measurement: focus on two things.

Behavioral consistency of the same config.yaml across the IDE extension and the CLI (cn) forms
The boundary of headless mode auto-excluding ask-permission tools — the official is explicit that "with nobody to approve, they are auto-excluded"; this is the easiest tripwire for scripted users.

当前状态：仓库已只读，不再更新⚠ Current status (the most important prerequisite)

继续用没问题，别再等更新。官网 continue.dev 现为收购公告，原文：

Continue has been acquired by Cursor... our open-source codebase remains freely available as a foundation for others.

仓库 README 顶部原文：

Note: The continuedev/continue repository is no longer actively maintained and is read-only for all users.

发布的最终版本是 VS Code 扩展、CLI 与 JetBrains 插件的「final 2.0.0 release」（releases 里可见 v2.0.0-vscode、v2.1.0-vscode）。

从 release note 看，最终版移除了 CLI 安装横幅与「Hub slug」，并加了 deprecation 提示；同时按 README 口径「removing anonymous telemetry」。

问题	结论
还能用吗	能，Apache-2.0 代码仍公开（仓库 read-only，归档日期官方未给出）
还会有更新吗	不会——README 明示「no longer actively maintained」
商业线还在吗	原 Hub 定价页已 404；continue.dev 域名下现为 Continuous AI

Keep using it if you like — just stop expecting updates. The official site continue.dev is now an acquisition notice, original wording:

Continue has been acquired by Cursor... our open-source codebase remains freely available as a foundation for others.

Repository README, top-of-file original wording:

Note: The continuedev/continue repository is no longer actively maintained and is read-only for all users.

The final released versions are the "final 2.0.0 release" of the VS Code extension, CLI and JetBrains plugin (v2.0.0-vscode, v2.1.0-vscode visible in releases).

Per the release notes, the final versions removed the CLI install banner and the "Hub slug", and added deprecation notices; per the README they also came "removing anonymous telemetry".

Question	Conclusion
Can it still be used?	Yes — the Apache-2.0 code remains public (repository read-only; the official gives no archival date)
Will there be updates?	No — the README states "no longer actively maintained"
Is the commercial line still around?	The old Hub pricing page is 404; under the continue.dev domain it is now Continuous AI

三种形态（README 与文档列出）Three forms (listed in the README and docs)

形态	关键能力
VS Code 扩展	Agent / Chat / Edit / Autocomplete 模式，工具需逐次批准
JetBrains 插件	同上；README 建议改用 CLI
CLI（`cn`）	终端 agent，TUI 与 headless 双模式，可进 CI / git hooks

Form	Key capabilities
VS Code extension	Agent / Chat / Edit / Autocomplete modes; tools need per-use approval
JetBrains plugin	Same as above; the README recommends the CLI instead
CLI (`cn`)	terminal agent, TUI and headless modes, usable in CI / git hooks

模型自配（官方列 30+ provider）Self-configured models (the official lists 30+ providers)

官方 model providers 页列出的 provider 覆盖：

Anthropic
Azure AI Foundry
Amazon Bedrock
DeepSeek
Gemini
Mistral
Ollama（本地模型）
OpenAI
Vertex AI
xAI

该页还指向「More（30 items）」。

配置即 config.yaml 里的 models，每条写 provider + model（+ 可选 apiBase），可指定 roles 与 capabilities（tool_use 是 Agent 模式的前提）。

这与 ./cursor.md 的差别最根本：Cursor 用其目录内模型， Continue 可以指任意 provider，包括完全本地跑。

Providers covered on the official model providers page:

Anthropic
Azure AI Foundry
Amazon Bedrock
DeepSeek
Gemini
Mistral
Ollama (local models)
OpenAI
Vertex AI
xAI

That page also points to "More (30 items)".

Configuration is the models block in config.yaml; each entry writes provider + model (+ optional apiBase), with roles and capabilities specifiable (tool_use is a prerequisite for Agent mode).

This is the most fundamental difference from ./cursor.md: Cursor uses models from its catalog, while Continue can point at any provider, including fully local runs.

MCP 与权限MCP and permissions

MCP 官方支持三种传输：stdio、sse、streamable-http；配置放在 .continue/mcpServers/，且可直接把 Claude / Cursor / Cline 的 JSON MCP 配置复制进来。官方一句硬限制要注意： "MCP can only be used in the agent mode"。

权限口径分两侧：

IDE：官方原文 "By default, Agent mode will ask permission when it wants to use a tool"，可用 tool policies 设为自动或排除；

CLI headless：官方原文「tools that would normally prompt for approval (ask permission) are automatically excluded」，要写文件必须显式 --allow Write（或 --allow "*"）。

MCP officially supports three transports: stdio, sse, streamable-http; configuration lives in .continue/mcpServers/, and JSON MCP configs from Claude / Cursor / Cline can be copied in directly. One hard official limit to note: "MCP can only be used in the agent mode".

Permission stance splits by side:

IDE: official wording "By default, Agent mode will ask permission when it wants to use a tool"; tool policies can set tools to automatic or excluded;

CLI headless: official wording "tools that would normally prompt for approval (ask permission) are automatically excluded"; to write files you must explicitly pass --allow Write (or --allow "*").

商业线：Continuous AI（与开源扩展分开读）Commercial line: Continuous AI (read separately from the open-source extension)

hub.continue.dev（2026-10-08 取到）现为 Continuous AI 的商业线，三档：

档位	要点
Starter	按量计费（pay-as-you-go）：创建并运行 agent，接 Slack / Sentry / Snyk
Team	按席位订阅（含每席 credits）：团队内共享私有 agent，Gmail/GitHub SSO
Company	定制：SAML/OIDC、BYOK、承诺用量 / 发票 / SLA

报价数字按本站的报价记录约定不存，见 hub.continue.dev/pricing。

注意这不是 IDE 扩展的订阅 —— 扩展本身免费开源。

这一商业线已转向 PR 检查为主的 Continuous AI，不再是「Continue Hub」。

hub.continue.dev (fetched 2026-10-08) is now the commercial line Continuous AI, three tiers:

Tier	Highlights
Starter	pay-as-you-go: create and run agents, hook up Slack / Sentry / Snyk
Team	per-seat subscription (credits per seat): share private agents across the team, Gmail/GitHub SSO
Company	custom: SAML/OIDC, BYOK, committed usage / invoicing / SLA

Price numbers are not stored per D9 policy; see hub.continue.dev/pricing.

Note this is not a subscription for the IDE extension — the extension itself is free and open source.

This commercial line has pivoted to Continuous AI centered on PR checks, no longer "Continue Hub".

核验说明Verification note

这份条目标为 confidence: partial，原因有三条：

已取到官方正文的：形态、模型自配、MCP、权限、许可与商业线结构。
已停止维护：上游仓库 read-only、官网被收购公告替换，未来不会再更新。
已无法从官方页取证的：Hub 旧价目、账户/订阅的去向等历史能力。

已核到的官方页：仓库 README 与 Releases、官网 continue.dev（含 www 变体）、文档站首页、MCP 深潜、Model providers、Agent quick-start、 CLI quickstart 与 CLI headless、商业线定价页 hub.continue.dev/pricing。

This entry is marked confidence: partial for three reasons:

Official body text obtained: forms, self-configured models, MCP, permissions, license and the commercial-line structure.
Maintenance has stopped: upstream repository read-only, official site replaced by the acquisition notice; no future updates.
No longer verifiable from official pages: old Hub pricing, the fate of accounts/subscriptions and other historical capabilities.

Official pages verified on 2026-10-08: repository README and Releases, official site continue.dev (incl. the www variant), docs-site homepage, MCP deep-dive, Model providers, Agent quick-start, CLI quickstart and CLI headless, commercial-line pricing page hub.continue.dev/pricing.