Edition #1 · 2026-06-05 · ranked by stars/day, not total stars · every link verified live at publish.
We don't tell you what's popular — popularity is lagging and gameable. We tell you what's *gaining speed right now*, and whether it's worth your attention. 46 live AI repos screened; the ones below earned the slot.
odysseus — ⭐57,047 · ↑11,409/day · Python
A self-hosted AI workspace that went from zero to 57k stars in five days — the steepest curve on GitHub right now, an order of magnitude faster than anything else this week. Worth watching purely for the velocity, but temper it: explosive day-one curves often include a launch spike. We're flagging it because the *acceleration* is real, not vouching for longevity yet.
Who needs it: anyone tracking where the self-hosted-AI-workspace category is heading.
claw-code — ⭐193,375 · ↑2,886/day · Rust
The fastest repo in history past 100k stars, and still climbing 3k/day at nearly 200k. A Rust-built agent runtime in the OpenClaw lineage. At this size it's a movement, not a project — if you're building agents, you need to know what it does and why everyone's installing it.
Who needs it: agent builders deciding what runtime to bet on.
antirez/ds4 — ⭐14,440 · ↑435/day · C
A DeepSeek-4-Flash local inference engine for Metal and CUDA, from Salvatore Sanfilippo (creator of Redis). Pedigree matters here: antirez writes famously clean, dependency-light C. If you run models locally, this is the most credible new inference engine of the week.
Who needs it: anyone doing on-device / self-hosted inference on Apple Silicon or NVIDIA.
mempalace — ⭐54,053 · ↑872/day · Python
Pitched as the best-*benchmarked* open-source AI memory system — and free. Agent memory is one of 2026's hottest unsolved problems; a benchmarked, open option is genuinely useful rather than another vibes-based memory wrapper.
Who needs it: anyone building agents that have to remember across sessions.
open-design — ⭐59,822 · ↑1,534/day · TypeScript
A local-first, open-source design tool (native desktop, web/mobile prototypes) positioned as a Claude-Design alternative. Local-first + open-source is the real hook — your design data stays yours.
Who needs it: designers and indie teams who want a non-cloud design surface for agents.
DeepSeek-Reasonix — ⭐18,685 · ↑406/day · Go
A DeepSeek-native terminal coding agent engineered around prefix-cache stability — a real performance idea, not a reskin. Notable as part of the fast-growing DeepSeek-native tooling wave.
Who needs it: terminal-first developers on DeepSeek models.
openai/codex-plugin-cc — ⭐20,342 · ↑304/day · JavaScript
Official OpenAI: use Codex *from inside* Claude Code to review code or delegate tasks. Signal worth noting — the labs are now openly building cross-tool bridges between each other's agents.
Who needs it: anyone running a multi-agent (Claude Code + Codex) coding setup.
A large share of this week's fastest-climbing repos aren't tools — they're skill packs for coding agents: gstack (⭐108k, Garry Tan's exact 23-tool setup), caveman (⭐69k, a skill that cuts ~65% of tokens by "talking like caveman"), graphify (⭐60k, code→knowledge-graph skill), career-ops (⭐49k, job-search skill). They're climbing fast and tell you where attention is — but most are configs/prompts, not infrastructure. We surface the wave as a *trend signal* and don't pad the tools list with them. (caveman's token-compression idea is the one with a reusable insight under the meme.)
Sourced live from the GitHub API (60 candidates, AI-filtered to 46), each verified not-archived and pushed within 30 days, ranked by stars/day since creation (cold-start velocity proxy — switches to 7-day star-delta once history accumulates), then curated for substance vs. noise. Star counts pulled at publish time — re-verify before reposting, they move daily.
*Built by the Autonomous AI Digest pipeline · catch acceleration, not stars.*