编辑甄选 · OSS
GitHub 高星项目· 3 projects
由潜龙编辑团队甄选,每周更新最值得看的开源项目。我们看重的不是 Star 多少,而是它解决问题的方式。
全部主题高潜项目aiai-agentai-agentsconnectordagfastapifunction-callingllmmcpnextjsopenai-compatiblepythonagent-observabilityanalyticsclaude-codeclicodexcoding-agentcost-trackingdashboarddeveloper-toolsdevtoolsagent-builderagent-orchestrationagent-workspaceagentic-aiai-automationai-coworkerai-skills-managerai-workflow-builderharnessagentic-biagentic-workflowagentsai-workflowsartificial-intelligencebusiness-analyticsbusiness-intelligencelangchainlanggraphragautomotivecardoipedge-aihome-assistantllama-cpplocal-llmobdoffline-aiagentic-codinganthropicclaudeclaude-code-alternativecppagentinvestmentlangraphskillstradingacpagentic-development-environmentcode-reviewcoding-agentsgithub-copilotgolangkanbanagent-securityai-agent-securityai-antivirusai-securitycursordevsecopsdsh-plugingemini-clichatbotcrmcustomer-supportecommercelgpdmulti-tenantnuvemshopopen-sourcechatgpthermeshermes-agentopenaidata-pipelinedata-processingmachine-learningsdktypescriptvscode-extensioncalendarfinancegdrivegmailgoogle-calendarnotionproductivityschedulershopifyai-codingastclinecode-intelligencecontext-windowcopilotcli-toolcodex-clideveloper-productivityhookslinuxllm-toolsagent-osai-coding-agentdeepseekdeepseek-harnessdshagent-skillsclaude-plugincomfyuicomfyui-extensioncomfyui-mcpcomfyui-mcp-serverfluximage-generationlocal-firstai-toolsbrowser-aibrowser-automationbrowser-extensionchrome-aichrome-devtoolschrome-extensionchrome-extensionsfirefox-addonfirefox-addonsagent-memoryai-infrastructureknowledge-graphllm-memorymcp-servermemory-managementmulti-agent-systemsagent-managementai-assistantai-workercross-platformdesktop-appdigital-personamodel-managementopenclawagent-skillclaude-code-pluginclaude-code-skillsclaude-skillscoding-assistantempryographdockerknowlege-graphn8nneo4jollamapgvectorci-cdclickhouseevalsevaluationllm-as-judgellm-evaluationllm-observabilityllm-opsai-sdkdecentralized-aillamallm-inferencelocal-ailocal-fine-tuningnodejson-device-aiopen-source-aiopen-source-llmopen-source-ragagentic-frameworkai-chatautomationautonomous-agentsgeminillm-agentsmcp-clientai-coding-assistantaiderbubble-teadiscordai-contentai-marketingawesome-claudecontent-creationcontent-engineeringlinkedin近半年高星agent-teamfimgaflutteropencodeopenpencilaccessibilityandroidautonomouskotlinmultimodal-aicall-graph-analysiscode-indexingcode-searchdeveloper-tools-ai-agentmcp-toolscybersecurityholehemaigretmodel-context-protocolosintrustterminaltuiai-developmentjavareactjsagile-workflowsclaude-aicode-analysiscodebase-auditdocumentation-generatoragile-scrumappledoubaominimaxparallel-agentsapi-gatewaycontinuegoosekilooh-my-piopenai-apichatgpt-alternativedifyflowisehomelabmake-alternativeelectronapisdata-flowdevelopmentlow-codeno-codebrowserbrowseroschromiumhacktoberfestlmstudiomacoswindowsagplai-agent-platformaudit-trailodoorbacself-hostedtelegram-botautonomous-pentestingbug-bountyethical-hackingpenetration-testingpentestpentesting-toolsreconggufagent-toolsai-workflowcommand-line-toolgomcts-rlmeta-agentsmeta-optimizationruntime-supervisiontree-rlworkflow-automationdemoeasyembeddingflaskgpthuggingface-transformersmultimodalqwencodecoderextensionidevscodecost-reductionllmopsmitm-proxypromptai-platformenterpriseclapembyjellyfinjellyfin-plugink3skuberneteslibrosalyrionmusicnavidromeagpl-v3ai-safetyauditingblockchaindcap-attestationdecentralized-marketplacehardhatintel-sgxpolygon-edgeapiembedded-ipaasintegrationsipaasanthropic-apiapple-silicondeepseek-v4diffusioninferencecode-understandingcodebase-searchgraph-databaseboltcypherdatabaseenterprise-solutionsgraph-raggraphqlhnswmemoryosmvccinstant-messaging
自
Q00
自我进化的 Agent OS
5.6k
这里有个小看点,Ouroboros 定位为一个 Agent OS,它的核心思路是让智能体在受控的环境下自我进化。它不像是在做一个普通的任务执行器,而是引入了分阶段评估和带有预算限制的进化循环。 它设定了严格的成功契约,评分指令和预期结果不会直接暴露给智能体。作为一个 MCP 服务器,它目前支持 13...
构
OpenBMB
构建复杂 RAG 流水线的低代码 MCP 框架
5.6k
UltraRAG 是由 OpenBMB 推出的一个低代码 MCP(Model Context Protocol)框架,专门用来构建复杂的 RAG(检索增强生成)流水线。 这个方向最近挺热,原因也不难理解:基础的 RAG 已经很难满足实际业务需求,开发者需要处理多模态数据、复杂的检索逻辑以及不同模型(...
基
pcliangx
基于 Claude Code 的研发团队脚手架
740
这个项目为 Claude Code 设计了一套非常重型的多智能体协作脚手架。它预设了 19 个不同的研发角色,通过阶段门禁、强制技能、阻塞钩子和完成标准(DoD)检查单来规范 AI 的工作流。它支持多模型接入,并且覆盖了 Web、微信小程序和苹果原生应用的开发链路。 别把它看成魔法棒,它更像是在把传...