The Lab

What to try


New software the crowd is converging on, measured — not recommended. Distinct people paying attention, ranked within each day, classified as AI tooling by model, with the classifier's rejects left visible.

Window opened Aug 12, 2026 · collectors run hourly and daily · generated Sep 15, 2026, 10:29 UTC

How this list is made

Each day, every starring and forking event GitHub’s public stream carries is counted as distinct people paying attention — people, never raw event counts, so one enthusiastic bot cannot manufacture a breakout. Repositories are ranked within their own day, the day’s floor is 5 people, and a language model classifies each arrival as AI tooling or not. Rows that failed that classification stay visible below, dimmed — you should see what the filter removed, not take the cut on trust.

Ranks are trustworthy; absolute counts are not. GitHub’s public stream heavily undersamples starring, and a fresh day is incomplete at read time — both are quantified on Trust, and both are why this page compares repositories only against the same day’s field.

step 170,566,903public GitHub events in thewindowstep 2distinct peopleone person counted once, notraw eventsstep 3ranked within its daynever against another day’sfieldstep 4floor of 5fewer people, no rowclassified: is this AI tooling?a language model, on every arrival16 rows, shownthe freshest day's field, ranked14 rows, dimmed — not deletedyou see what the filter removed

Ranking inside a single day is the load-bearing choice, marked here in red. GitHub’s public stream undersamples starring badly and a fresh day is still filling in, so a count from Tuesday and a count from Saturday are not the same unit — but position within a day’s own field is.

Every step is a rule, not a judgment. The last one is a language model, so its rejects stay on the page — 14 of the 30 rows in the freshest day are dimmed rather than deleted.

Window = where the repo ranked in each day’s field, Aug 12Sep 15; top of the line is top of GitHub. A gap is a day the repo drew nothing in the sample; a line that begins mid-window is a repo that did not exist before it.

Sep 15, 2026fresh — this day is still filling in; ranks hold, totals will rise

#RepositoryWindowIn sampleWhat it is
1alibaba/open-code-reviewcode-reviewreliability134Fast, efficient, battle-tested at Alibaba's scale. Hybrid architecture code review tool: deterministic pipelines + LLM Agent, precise line-level comments, built
2deepseek-ai/deepseek-harnessagent-harnesstool-use101DeepSeek Harness: Everything is a Plugin.
3deeplethe/utopiamemoryrag93World's first open-source enterprise world model.
4debpalash/VoiceStudiolocal-modelsprivacy92VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation i
5bojieli/ai-infra-booknot AI tooling85《深入理解 AI Infra:量化分析与系统设计》(李博杰 著)开源书稿:从硬件约束和模型架构出发,量化推导 LLM 推理与训练系统设计。含全书正文、PDF、配套计算工具与实验
6TencentCloud/Octopagent-harnessmemory81A smarter, self-hosted AI assistant — multi-user, multi-agent.
6hypit-ai/hypitnot AI tooling81Clone any viral video with AI agents. Not just a script, the whole workflow: swap the face, the words, the B-roll, ship 100 variants in one command, and get you
8JustVugg/colibrilocal-modelsinference-cost80Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
9Panniantong/Agent-Reachagent-native-accesstool-use76Give your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
10tt-a1i/archifydesign-toolingcode-generation74Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
11bilawalsidhu/gods-eye-viewgeospatial-visualizationosint-intelligencenot AI tooling65A spy satellite simulator in your browser, except the data is real. Live open source spatial intelligence on a photorealistic 3D globe.
12abue-ammar/tinycastdesktop-uxnot AI tooling62Tinycast — a tiny, fully native macOS launcher, hotkeys, and clipboard history. CA: 682r79Fjc3U2aMEZASFSbd5SW2u4bzqfVPAHofsipump

Sep 14, 2026fresh — this day is still filling in; ranks hold, totals will rise

#RepositoryWindowIn sampleWhat it is
1JustVugg/colibrilocal-modelsinference-cost78Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
2debpalash/VoiceStudiolocal-modelsprivacy75VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation i
3shadcn-ui/lintnot AI tooling72An agent-first linter for Tailwind design systems. Write design system rules that agents can verify.
4alibaba/open-code-reviewcode-reviewreliability69Fast, efficient, battle-tested at Alibaba's scale. Hybrid architecture code review tool: deterministic pipelines + LLM Agent, precise line-level comments, built
4bilawalsidhu/gods-eye-viewgeospatial-visualizationosint-intelligencenot AI tooling69A spy satellite simulator in your browser, except the data is real. Live open source spatial intelligence on a photorealistic 3D globe.
6bojieli/ai-infra-booknot AI tooling63《深入理解 AI Infra:量化分析与系统设计》(李博杰 著)开源书稿:从硬件约束和模型架构出发,量化推导 LLM 推理与训练系统设计。含全书正文、PDF、配套计算工具与实验
7cloudflare/security-audit-skillnot AI tooling56A coding-agent skill for multi-phase security audits with independently verified, machine-readable findings
8deepseek-ai/deepseek-harnessagent-harnesstool-use52DeepSeek Harness: Everything is a Plugin.
8tt-a1i/archifydesign-toolingcode-generation52Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
10DietrichGebert/ponytailcode-generationcontext-engineering43Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
11TauricResearch/TradingAgentsdata-analysisrole-specialization40TradingAgents: Multi-Agents LLM Financial Trading Framework
11k1tbyte/Wand-Enhancerux-enhancementnot AI tooling40Advanced UX and interoperability extension for Wand (WeMod) app

Sep 13, 2026fresh — this day is still filling in; ranks hold, totals will rise

#RepositoryWindowIn sampleWhat it is
1debpalash/VoiceStudiolocal-modelsprivacy414VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation i
2bilawalsidhu/gods-eye-viewgeospatial-visualizationosint-intelligencenot AI tooling382A spy satellite simulator in your browser, except the data is real. Live open source spatial intelligence on a photorealistic 3D globe.
3finlionanger/AutoCad-setupnot AI tooling258no description on GitHub
4InletBackerTier/SolidWorks-for-Windowsnot AI tooling256no description on GitHub
5decioriolepartition/qgkknlvfnot AI tooling248no description on GitHub
6ScalarLinkAxe/Optimizer-Toolkitnot AI tooling246no description on GitHub
7HeartNautilusGuide/KMS-Piconot AI tooling229no description on GitHub
8google/artemisnot AI tooling226ARTEMIS turns natural-language instructions into reliable Android automation. It automates end-to-end workflows, captures logs for debugging and analysis, and i
9deepseek-ai/deepseek-harnessagent-harnesstool-use219DeepSeek Harness: Everything is a Plugin.
10JustVugg/colibrilocal-modelsinference-cost211Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
11sanhaji182/financial_tracker_plannernot AI tooling206no description on GitHub
12tt-a1i/archifydesign-toolingcode-generation186Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.

New on the watch this week

2075 repositories crossed the attention floor and joined the watch in the last seven days — anything that crosses it gets picked up automatically and snapshotted daily from then on. (The watch itself was seeded Aug 20, 2026 with the monthly instrument’s 2,142 tracked repositories; these arrivals are what the daily instrument has added on top.) Recent arrivals that classified as AI tooling:

Would it have caught the last breakout?

Before this lane went live it was run backward over its own window. deepseek-ai/deepseek-harness was created Aug 13, 2026 and ranked #2 on all of GitHub by attention in its first hour alive, and #1 for 7 of its first eight hours. On its peak day (Aug 14) it drew 478 people in the sampled stream against 17 for the runner-up — a 28× gap. The runner-up that day, cordiverse/cordis, is the framework underneath it: the crowd went looking at what the new thing was built on, and the instrument caught that too.

The harness shipped with its issue tracker disabled — a contribution-based sensor would have missed it entirely. What fired instead was the satellite pattern: the framework underneath it (+1h), the plugin index (+2h), a web UI (+3.5h), a desktop app (+9.5h) all drew their first attention within hours of it. Satellite clustering is the shape of a breakout, and it is the only signal when the artifact itself arrives closed.