GitHub listed seventeen projects this Monday morning. The lead is a security tool for once: AiSOC, a self-hosted AI security operations centre that keeps a human approver in charge of responses. Claude-mem returns, not for its first place but because its new release admits that its installer uploaded 30 days of prompts without asking. The rest is practical: Handy for offline dictation, FreeToken for big local models, Kornia's audited 0.9.0 and Hyprland 0.57.0. Curation order reflects practical interest, not GitHub rank. Trending is not a release date or an endorsement.
Picks are listed in curation order, not GitHub’s rank; each card shows the rank GitHub displayed when the page was checked. Stars measure attention, not security, quality or fitness for use, and no project here was installed or run. “Stars today” is GitHub’s own displayed figure for its daily window, not an audited rolling 24-hour count.
What it does. A self-hosted AI security operations centre. It takes in security telemetry from pushed events or 90 data connectors (Splunk, Sentinel, Elastic, CrowdStrike, Okta, AWS and others), runs detection rules, groups alerts into incidents and has an AI agent investigate each one. Every prompt, tool call and verdict goes into an Investigation Ledger that can be exported as a signed evidence bundle, and response actions are proposed for a human to approve.
Caveat. The core profile needs Docker with about 8 GB of memory and 20 GB of disk and downloads a ~2 GB local model (Llama 3.2 3B) that is, in the README's words, not a frontier model. It is largely one maintainer's work (about 1,050 commits, far more than any other person) and reached v21.0.0 in six months, with a breaking change to alert grouping in that release. Its benchmark figures are the maintainer's own, and it is not a drop-in SIEM replacement. For authorized monitoring only; we did not run it.
Attention, not endorsement
GitHub showed
“446 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
What it does. A persistent memory plugin for coding agents (Claude Code, Codex, OpenCode, Gemini and others). Hooks record what the agent does in a session, a model compresses it into summaries kept in local SQLite with vector search, and later sessions get relevant context back. An optional cmem.ai cloud service adds a hosted summarizer and sync.
Caveat. The release notes say prompts already uploaded are not removed by this update, and server-side handling at cmem.ai is outside the fix; the issue that reported it (5176) was still open when we checked. Anyone who ran the installer's browser setup between v15.2.0 and v15.8.2 should assume that history was sent. It still captures everything your agent does, so secrets in tool output can reach its database. We did not install or run it.
Attention, not endorsement
GitHub showed
“1,526 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
What it does. A free, offline speech-to-text app for Windows, macOS and Linux. Hold or tap a shortcut, speak, and the transcription is pasted into whatever app has focus. Silence is filtered with Silero voice detection, and you choose the speech model: Whisper (Small to Large, with GPU acceleration where available) or the CPU-friendly Parakeet V3 with automatic language detection.
Caveat. Still pre-1.0, and the README lists known issues: Bluetooth headset microphones can degrade audio on macOS, and fn/Globe-key shortcuts work only on Apple keyboards. It needs microphone and accessibility permissions to paste into other apps, and the app downloads its speech models, which can be large, before it can transcribe (a manual route exists for restricted networks). The Homebrew and winget packages are not maintained by the Handy developers. Transcription accuracy depends on the model you pick; we did not test it.
Attention, not endorsement
GitHub showed
“339 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
What it does. An engine for serving large Mixture-of-Experts models on consumer GPUs. It splits work between the GPU and CPU, keeps recently used experts cached in VRAM and reuses saved context when an agent edits its history, and offers OpenAI- and Anthropic-compatible APIs so coding agents such as Codex, Claude Code and OpenCode can use a local model. A desktop app for Windows and Linux is offered from its website.
Caveat. The README's "290B+ models at blistering interactive speeds" is a maintainer claim we did not measure. The documented install needs Linux x86_64, an NVIDIA GPU (RTX 30 series or newer) with driver r580+ and a CUDA 13 toolkit with nvcc, since kernels are compiled on first use; AMD support is marked work in progress. Model weights are large downloads with their own licences, and the docs include a guide for repairing old checkpoints in its own weight format. We did not run it.
Attention, not endorsement
GitHub showed
“242 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
What it does. A differentiable computer vision library for PyTorch: image filtering, augmentations, geometry, feature matching (SIFT, RANSAC) that run batched on the GPU and pass gradients through, so they can sit inside a training pipeline.
Caveat. The speed-ups in the release notes (for example 16x faster Gaussian blur on Apple CPUs, faster than OpenCV on CUDA) are maintainer benchmarks we did not reproduce; the notes themselves say it is about 4x slower than OpenCV on CPU. Removing broken models such as SAM3 can break code that imported them, so pin the version and read the changelog before upgrading. It needs PyTorch, which is a large install.
Attention, not endorsement
GitHub showed
“101 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
What it does. An independent tiling Wayland compositor for Linux (not built on wlroots, KWin or Mutter) with animations, blur, plugins, socket IPC and dwindle, master and scrolling layouts. Its configuration reloads as soon as you save it, and recent releases move it to a Lua configuration API.
Caveat. Linux only, and it moves fast. The 0.57.0 notes list removal of legacy config support, a deprecated option, two options moved to a new category and new deprecation notices, so check your config and plugins before upgrading. A tiling compositor takes time to set up compared with a ready-made desktop, and BSD-3-Clause licensing covers the compositor, not every plugin. We did not install it.
Attention, not endorsement
GitHub showed
“38 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
Every repository GitHub listed, in GitHub’s order, with the figures it displayed when the page was checked (): 6 picked today, 7 covered in an earlier digest, 4 not covered. For projects covered before, growth compares the total stars at our first pick with today’s displayed total.
TypeScript · 1,526 stars today · 109,800 stars in total
Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
Picked today — Pick 2 above; first covered 30 Oct 2026 at 105,064 stars; now 109,800: +4,736 stars (+4.5%) in 3 days, about 1,579 a day.
Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
TypeScript · 175 stars today · 9,697 stars in total
Open-source AI coworkers that each get a computer of their own: a browser, files and tools, with every action decided before it happens and recorded after. Bring any AG-UI agent.
Covered before — first picked 30 Oct 2026 at 8,329 stars; now 9,697: +1,368 stars (+16.4%) in 3 days, about 456 a day.
Every project picked in an earlier digest, biggest growth first. “Then” is the total at our first pick’s metadata check; “now” is GitHub’s API star count checked . Growth measures attention, not quality or safety.
All 17 listing entries were inspected. Exact primary public GitHub repository metadata was checked for the nine entries not featured before and for every pick between 07:00 and 07:03 UTC, with README files, releases and licences for the picks; all were active and unarchived, and every pick has an API-detected licence.
thedotmack/claude-mem is a repeat (first featured 30 October). It returns only because of release v15.8.3 (2 November), a privacy fix described in its own release notes and issue 5176; its card says what changed.
Seven entries featured before were left out without a material change: lightpanda-io/browser and pingdotgg/t3code (31 October), milind-soni/OpenMausBot, Donchitos/Claude-Code-Game-Studios and CopilotKit/OpenBot (30 October), Renhuai123/ziwei-doushu (24 October) and shiyu-coder/Kronos (20 October). Their release lists show only nightly or patch builds since their cards, or releases that predate them.
Not picked: togg53192-cmd/jailbreaks, a list of prompts for bypassing AI model safeguards with no licence; public-apis/public-apis, a mature directory of free APIs; patchy631/ai-engineering-hub, a collection of AI tutorials that promotes its authors' newsletter; and anomalyco/hex, a newer local dictation app with a short README that overlaps Handy.
The preceding seven calendar dates, 26 October-1 November, were checked for local history; all seven editions exist.
The official listing was captured at 08:00:34 Europe/Paris. Literal stars-today labels are GitHub's displayed period figures; API total stars and README evidence were checked separately at the times shown per card. They are not an independently audited rolling 24-hour growth measurement.
Feature, performance and privacy descriptions are the maintainers' own, including AiSOC's rule counts and benchmarks, FreeToken's speed claims and Kornia's release benchmarks. No candidate code, installer, plugin or model was installed, downloaded or run, and none of this is a security audit or endorsement.