Devika AI开源 AI 软件工程师，能够从自然语言目标进行研究、规划并编写代码。

4.2 (5)

审阅者 Daniel Nikulshyn·更新 2026年5月

Open Source AI Agent Developer Tools Code Generation Self-Hosted LLM Research Automation

概览

Devika AI 是一款开源的 agentic 开发者助理，旨在接受高层指令并将其转化为可运行的代码。它会将任务拆解为多个步骤，浏览网页获取相关信息，并在多种编程语言中编写实现，目标是作为协作式的初级工程师，而非单纯的自动补全工具。该项目定位为社区驱动的替代方案，以区别于 Devin 等闭源工具，支持包括本地模型在内的多种大语言模型（LLM）。开发者可以自行托管 Devika，审查其推理过程，并扩展其行为，非常适合需要透明度和对 AI 代理与代码库交互方式进行控制的团队。

主要功能

自然语言任务分解与规划
自主网页浏览以进行研究
多语言代码生成
可插拔 LLM 支持（OpenAI、Anthropic、本地模型）
内置聊天和项目工作区
可自托管，拥有开源代码库

价格

模型: Freemium
分类: AI Agents
评分: 4.2 / 5 (5)

使用场景

根据纯英文规格原型化功能

开发者用自然语言描述期望的功能，让 Devika 规划步骤、研究 API，并生成多文件代码作为起始实现。

私有团队自托管 AI 编码

需要数据控制的团队可以在自有基础设施上部署 Devika，使用本地 LLM，确保源代码和提示不泄露至第三方服务器。

基于研究的自动化编码任务

利用 Devika 的自主网页浏览收集文档、示例和参考资料，并直接将这些发现整合进生成的代码中。

透明的闭源代理替代方案

采用 Devika 取代 Devin 等专有工具，以检查代理推理、审计步骤，并通过其开源代码库扩展行为。

优点 & 缺点

优点

免费且开源，可自行托管
支持多种 LLM 后端，包括本地模型
在任务规划中进行网页研究
透明的代理推理和步骤追踪

缺点

部署设置相较于托管方案更具技术门槛
输出质量高度依赖所选 LLM
仍在持续迭代，可能在生产环境中不够稳定

评测

4.2

5 个评分的平均值。

登录以留下评测。

Ahmed Saleh

May 13, 2026

Compared a few options

Evaluated this against two competitors. Where it wins: built-in chat and project workspace and performs web research as part of task planning. On balance the feature set — especially self-hostable with open-source codebase — justifies the 5 stars for our use case.

Victor Nguyen

Jan 2, 2026

Solid for our team

We rolled this out across the team last quarter and free and open-source with self-hosting. Self-hostable with open-source codebase fits neatly into how we already work, and self-hostable with open-source codebase removed a step we used to do by hand. Still evolving and may be unstable for production use, which is the main caveat, but it has held up under daily use.

Beatriz Costa

Oct 21, 2025

Solid for our team

We rolled this out across the team last quarter and supports multiple LLM backends, including local models. Natural-language task decomposition and planning fits neatly into how we already work, and self-hostable with open-source codebase removed a step we used to do by hand. Still evolving and may be unstable for production use, which is the main caveat, but it has held up under daily use.

Marcus Bell

Oct 4, 2025

Does the job

Pretty happy overall. Autonomous web browsing for research just works and performs web research as part of task planning. Setup is more technical than hosted alternatives can be annoying, but no dealbreakers — I'd recommend it to a friend without hesitating.

Olga Ivanova

Jun 3, 2025

Solid for our team

We rolled this out across the team last quarter and performs web research as part of task planning. Natural-language task decomposition and planning fits neatly into how we already work, and built-in chat and project workspace removed a step we used to do by hand. Setup is more technical than hosted alternatives, which is the main caveat, but it has held up under daily use.