Firehose

October 2026

Public projects created in October 2026, day by day: GitHub repositories, datasets and AI models.

31 October 2026

New repositories covered agent tooling and dictation cleanup, with an agent plugin for capturing mid-thread ideas into in-repo sprints and a catalogue of LoRA format adapters on a small dictation model. Two further repositories documented conversational AI triage and batch inference mechanics on OpenRouter. On Hugging Face, the new dataset was a reference list of OpenRouter batch-capable models with batch and sync prices.

GitHub Repositories: 4 · Datasets: 1

30 October 2026

A command-line tool for spoken-word audio applies silence and filler-word removal, pause shortening, and voice-fitted equalization to reach a target loudness. The work sits in podcast post-production and audio processing.

GitHub Repositories: 1

29 October 2026

New repositories covered voice-data preparation and rclone administration, with a tool that trims clicks from training recordings and agent skills for remotes, mounts and config backup. On Hugging Face, the work was wake-word and dictation cleanup: two "Hey Yossi" models, a fine-tune that edits raw dictation, and a reference dataset.

GitHub Repositories: 2 · Datasets: 1 · AI Models: 3

28 October 2026

New repositories covered agent tooling and image-model evaluation, with a plugin for documenting and analysing home noise pollution and a periodic check of whether image models can produce stylised tech diagrams with correct text and arrows. On Hugging Face, the new work was evaluation material for that graphics benchmark: its test prompts, every generated image, and the raw human ratings.

GitHub Repositories: 2 · Datasets: 3

27 October 2026

Agentic Dev Workflows documents how to build software with AI agents across platforms, while a bounded brainstorming fine-tune generates novel items matching a format without repeating priors. An Android selection tool converts text case on device, and a personal speech-recognition model handles English dictation with tech vocabulary.

GitHub Repositories: 4 · AI Models: 1

26 October 2026

A dataset of real recordings of Jewish and Israeli first-name wake words, for training local wake word models with openWakeWord and microWakeWord, for Home Assistant Assist and desktop use.

Datasets: 1

25 October 2026

Agent plugins for software comparison, Israel Post services and Voicenotes note access sat alongside tools for predicting speech-recognition errors and restoring crashed terminal sessions.

GitHub Repositories: 5

24 October 2026

Agent tooling covers a Kaizen plugin for steady task breakdown and a hook that spots unrelated tasks and moves them to a fresh session. Speech recognition work includes a model that adds paragraph breaks to flat transcripts, an ASR cleanup model, and training data for both.

GitHub Repositories: 3 · Datasets: 3 · AI Models: 2

23 October 2026

Two agent plugins for Ubuntu cover machine administration and host security. The first handles health, services, logs, packages, storage, networking, hardware, Docker and KDE on desktops or servers, locally or over SSH. The second covers firewall rules, antivirus and rootkit scanning, and security posture audits.

GitHub Repositories: 2

22 October 2026

Agent plugins for Claude Code covered microphone tuning, music video generation, planning documents, spending limits, Ubuntu hotkeys and vendor bug reporting. Election work included a 2026 Knesset model and reference datasets on the outgoing Knesset, candidate lists and polls. A voice-clone conversation and its alignment data rounded out the repositories.

GitHub Repositories: 8 · Datasets: 3

21 October 2026

Claude Code plugins for Bluetooth control on Linux, temporary network overflow, spam filtering, Maccabi Online skills, and a marketplace guide. Small text models handle paragraph breaks in speech-to-text output, bolding key lines in dictated emails, and swapping identifying details for substitutes.

GitHub Repositories: 10 · Datasets: 2 · AI Models: 4

20 October 2026

Android app work covered a thin fork template with weekly upstream merges and signed releases, a public F-Droid repository, and an offline family first-aid app for Israel. Other repositories hid encrypted messages in spoken sentences. Datasets graded Israel-Palestine framing outputs and labelled unsolicited outreach emails.

GitHub Repositories: 4 · Datasets: 2

19 October 2026

Model evaluations dominated the work, covering political framing, propaganda bait, conspiracy endorsement, American centrism and moralising, with matching prompt datasets and raw run outputs. Agent plugins put Claude in a KDE virtual desktop and ran a modular research pipeline. Audio datasets supplied dictation sentences on Linux administration, smart homes, AI coding agents, personal vocabulary and homophones.

GitHub Repositories: 12 · Datasets: 13

18 October 2026

Two tools cover automated political decision making and long term archival storage. One is a constrained model that votes for, against or abstains on parliamentary motions with a stated reason for a declared constituency. The other is a self hosted web interface that retrieves archives from cloud storage and burns them to M-DISC BD-R discs.

GitHub Repositories: 2

17 October 2026

Two audio datasets cover read-aloud English speech: one of plain everyday sentences with homophone pairs, and a small hand-picked set of real voice notes.

Datasets: 2

16 October 2026

Agent tooling covers a per-turn context router, ChatGPT plugins, and skills for Israeli research and right-to-left work. Text classifiers sort voice-note transcripts into tasks and separate genuine stranger email from pseudo-personalised outreach. A dictation cleanup model and its evaluation dataset are planned.

GitHub Repositories: 7 · Datasets: 1

15 October 2026

Indexes of English-Hebrew code-switching work and of Hugging Face releases, plus a directory of AI activity in Jerusalem, sit alongside an agent plugin for planning custom small models. New datasets cover household inventory specification lines and voice-note transcripts, and fine-tuned models classify Hebrew words in Latin letters, voice-note intent, and inventory specification attributes.

GitHub Repositories: 4 · Datasets: 2 · AI Models: 3

14 October 2026

Regex snippets for cleaning up speech-to-text output, built for Android first with Linux engine differences flagged, sit alongside a dictation-style English audio dataset of sentences dense in software, DevOps, cloud, self-hosting and AI vocabulary.

GitHub Repositories: 1 · Datasets: 1

8 October 2026

A Claude Code plugin for kaizen, the incremental improvement of mature codebases one atomic commit at a time.

GitHub Repositories: 1