Tag
Explore by tags
A 26M-parameter LLM distilled for reliable function-call generation on tiny devices, with open weights, local finetuning tooling, and a web playground for on-device testing. Pretrained at scale then post-trained on a single-shot function-call dataset for tool integration.
One-command installer that gives AI agents the ability to read and search the web and social platforms (web pages, Twitter/X, Reddit, YouTube, GitHub, Bilibili, XiaohongShu) by installing and wiring upstream CLIs and MCP connectors while keeping credentials local.
Provides a deployable personal AI assistant that runs locally or in the cloud, supports multi-channel chat, extensible Skills/Plugins, and local-model runtimes. Key features include three-layer memory, kernel-level sandboxing and tool/file guards, and bundled QwenPaw-Flash local models for zero-API deployments.
Runs local AI models on Apple Silicon as an OpenAI‑compatible server, emphasizing low latency, prompt caching, and reliable tool-calling. Optimized for M1–M4 Macs with multimodal support and drop‑in compatibility for IDEs and agent frameworks.
Unmixes green‑screen pixels with a neural model to recover straight (unmultiplied) foreground color and a clean linear alpha for every pixel, preserving hair, motion blur and translucency. Produces VFX‑standard EXR outputs, supports optional AlphaHint generators (GVM/VideoMaMa) and Docker/consumer‑GPU optimizations.
Builds a local structural knowledge graph of a codebase so AI coding assistants read only the minimal, relevant code during reviews and daily tasks—reducing tokens used while providing blast-radius impact analysis, incremental updates, and MCP integrations.
Audits and reduces token waste in LLM sessions by compressing verbose outputs, checkpointing before compaction, and restoring lost context. Runs fully locally with zero telemetry and provides a live token dashboard plus plugins for Claude Code, OpenClaw and Codex.
Local-first desktop workbench that scrapes job leads, filters low-quality postings, scores candidate fit with explainable rules and vector matching, and generates tailored resumes, cover letters, and outreach drafts while keeping data on-device.
Extracts derived keys from running WeChat 4.x processes to decrypt SQLCipher 4 databases and .dat media files, and provides a real-time message monitor with a Web UI. Cross-platform (Windows/Linux/macOS) but requires process-memory or local-data access and is intended for decrypting your own WeChat data only.
Performs deterministic, sub-millisecond policy checks on every agent action (allow/deny + audit) and adds zero-trust identity, execution sandboxing, and SRE features. Covers the OWASP Agentic Top 10 and is designed to sit between agent frameworks and runtime actions for auditable, low-latency governance in production.
Identifies and surgically removes the internal activation directions that trigger refusal behavior in large language models, with one-click options on a HuggingFace Space or a local Python API. Combines multiple extraction methods (SVD, whitened SVD, sparse autoencoders), reversible steering, and analysis-informed verification to quantify capability and refusal trade-offs.
