编辑甄选 · OSS
GitHub 高星项目· 2 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agent-skillsagentsai-agentsai-toolsanthropicchatgptclaudeclaude-codeclaude-skillscursordeveloper-toolsgeminiaiai-gatewayapi-key-managementbudgetscost-trackinggatewaylitellm-alternativellmllm-gatewayllm-proxyllm-routingagent-securityagentic-securityaudit-logcausal-proof-enginellm-securitylocal-firstmcp-securityagentic-aiai-agentai-codingai-coding-assistantautonomous-agentsclicode-assistantcoding-agentagents-mdcodexllm-toolsmcppm-toolsproduct-managementacp-clientacp-gatewayagentic-workflowai-memoryclaude-plugindsh-plugingemini-cli-extensionhermes-agentdigital-twinknowledge-graphmemorymulti-agentpersonal-airagagent-teamsai-automationautonomous-agentawsbusiness-operationsclaude-code-plugindevopsecommerceagent-infrastructureagent-platformagent-runtimeagents-apiai-infrastructurecloud-agentsdurable-agentsdurable-executionmanaged-agentsai-assistantlangchainlangchain-pythonlocal-llmollamaprivacyreact-agentsovereign-aiandroidcomputer-usecross-devicedesktop-appgui-agentioslinuxmeta-museai-methodsapidockerfastapimthdspipelexa2aagent-evaluationllm-agentsllm-evaluationmodel-context-protocolpythonreinforcement-learningreinforcement-learning-environmentssandboxagentcode-generationcode-intelligencecompilercross-platformharnessharness-engineeringno-codeopenaiorchestrationxinferenceagent-swarmautomationbuncode-reviewgithubguardrailsai-workstationcomfyuilocal-aimcp-serverai-pentestingattack-surface-managementbug-bountycybersecurityethical-hackingethical-hacking-toolsexploitationkali-linuxagent-orchestrationagentic-workflowsai-software-engineeringgoreactself-hostedsoftware-factoryworkflow-automationcorecoderdeepseekapple-siliconinference-enginellm-inferencemacosmetalmixture-of-expertsmlxmoeframeworkgoogle-antigravitygraphragknowledge-managementcheatingcluelycluely-alternativecluely-clonefinal-round-aiagent-frameworkartificial-intelligenceopen-sourcerusthacshome-assistanthome-automationai-hedge-fundai-quantbyokdocker-composefinancial-aifintechinvestment-researchllm-agentccoding-assistantllama-cppopenrouterreplagentic-codingai-governancecontext-engineeringcontext-gatewaycontext-intelligencecopilotcsharpcudacuda-programmingdotnetdsparkgemmagemma4ggufglminferenceanalyticsantigravitycli-toolcursor-aiinsightskirolarge-language-modelsai-workbenchdata-analysisdeepseek-harnessdshelectronbatch-processingcontent-extractiondata-collectionfirecrawlfirecrawl-aijavascript-renderingagent-client-protocolcodex-clicourse-downloaderdownload-managerdownloaderhotmart-downloaderinstagram-downloaderai-note-takingai-second-brainclaude-code-skillclaude-memoryclaude-skillkarpathy-llm-wikiai-orchestrationbusiness-automationenterprise-aierpglobali18npipi-agentpi-desktoppluginsai-platformchatbotconversational-aigenerative-aillmopsoperating-systemclaude-code-skillsmulti-modelruntimeai-identitydigital-soulgemini-clilong-term-memoryagent-memorycopy-on-writeembedded-databasefull-text-searchhnswhybrid-searchllamaindexmysqloceanbaseai-sdkhuman-in-the-loopa2a-protocolagent-cliai-agents-cliai-workforcellm-runtimelocal-ai-agentsagent-harnessagent-observabilityagent-pluginagent-sandboxaudit-loggingamdgfx1151gpu-inferencehipllm-servinglong-contextopenai-apiqwenacrylagent-context-relayagenticagentic-development-environmentai-agent-permissionsai-safetyapi-gatewayaudit-trailauthorizationdecionisexecution-authorityagent-workspaceapple-intelligencecppdiffusion-modelsedgeflutterkotlinllamacppmultimodalelectron-appguiembeddingsknowledge-basesemantic-searchagent-osgrok-bot-alternativemuse-alternativeagent-loopsdata-catalogdata-governancedata-integrationdata-lineagedata-pipelinedata-platformdatabase-managementer-diagraminternal-toolsdatabase-clientdatabase-guidatabase-toolmariadbpostgresqlai-animationai-video-generatorcapcutclaude-agent-sdkcontent-creationffmpegimage-to-videomulti-providernovel-to-videomarkdownmuapibm25coding-agentsgit-nativegolangllm-memoryai-code-reviewdevtools
A
peonist-ai
AMD Strix Halo 专属推理优化
617
这个项目非常底层且硬核,专门针对 AMD 的 Strix Halo(gfx1151)架构优化了大语言模型的推理速度。它的核心目标很明确:提供在这款特定 AMD 硬件上运行 Qwen3.8-Flash-Next 模型的最快方案。 这里有个小看点,目前开源社区对 Nvidia 生态的优化已经非常成熟,但...
在
carloslfu
在 Mac 上流式运行 105GB 大模型
413
这个项目的思路非常巧妙:它试图让内存有限的 Mac 也能跑起 105GB 的庞大模型。具体来说,它通过从 SSD 流式加载 Qwen3.8-Flash-Next(一个 125B 的混合专家模型),并仅仅将最活跃的专家网络缓存到内存中,从而使得 16GB 到 64GB 内存的 Mac 也能完成推理。...