Featured
New 2026

Voice-note transcription for Android - bring your own API key. One-tap recording, voice keyboard (IME), LLM transcript cleanup, local history.

New 2026

Geometric Layer Profiling for Training-Free LLM Pruning — NeurIPS 2026 workshop submission

New 2026

Scout, a self-hosted Discord memory bot. Listens to the conversation and creates summaries and answers questions

New 2026

a hobby proejct for evading pangram detection

New 2026

Pinned one-DGX-Spark Docker recipe for DeepSeek V4 Flash with EXL3, SparkInfer, and 262K NVFP4 MLA KV cache

New 2026

活人感写作 1.0.0|通用中文创作与改稿 Skill

New 2026

Open source AI resume tailor. Score your resume against any job description, tailor it without invented experience, and export a PDF in ten templates. Runs on your own LLM API key, no account, nothing stored on a server.

New 2026

Cohere Transcribe + VAD, forced alignment, and diarization for accurate, timestamped, speaker-labeled transcripts.

New 2026

拆开 GPT系列

New 2026

VENEER: how much of an LLM judge's verdict is formatting, not substance? Same claims, nine renderings, 972 pairwise judgments.

New 2026

LocusTether is a self-hosted replacement for the Omi wearable's normal cloud pipeline. Omi's official app streams your audio to Omi's servers, where a cloud LLM turns it into summaries, to-do lists, and searchable notes. LocusTether gets the same value without any of that audio or text ever leaving your own hardware.

New 2026

Synesthesia Encoder (通感编码器) — give any text-only LLM (DeepSeek, etc.) the ability to see images via structured spatial text encoding. A Pi agent extension.

New 2026

Mergeable, privacy-safe sketch primitives for LLM/GenAI telemetry - HLL++, weighted frequent-items, Bloom, MinHash. Domain-separated keyed hashing, byte-deterministic wire format, conformant pure-Go and pure-Python implementations.