A developer has released Custodian, an open-source (Apache 2.0) Windows desktop assistant that combines Qwen open-weight models, local inference via node-llama-cpp, a Tauri/React interface, and a Rust…
A developer built Voice Order Taker, a local Streamlit app that transcribes WhatsApp voice notes in Urdu, Roman Urdu, Pashto and English and extracts them into a structured order sheet using Whisper a…
A developer has published sizing math showing that vLLM's fp8 KV-cache flag halves KV-cache memory, letting a Llama 3.1 8B model with a 64K context fit on a 24 GB RTX 4090 at roughly $0.35/hr instead …
An engineer built Proveniq, an open-source AI-assisted asset provenance and license-screening tool that uses Gemma to extract license details while keeping the model from inventing evidence or making …
A developer built QuoteFlow, an open-source tool that converts voice or text jobsite notes into itemized, region-aware contractor estimates with materials, labor, permits, tax and a branded PDF. The a…
A developer built Google Photos + AI Storage Assistant, a prototype photo-gallery web app with albums, categories, favorites, an album-suggestion interface, and a local AI assistant powered by Ollama …
An engineer proposes replacing single-context LLM agents with a four-tier multi-agent topology — Scout, Architect, Engineer, and Sentinel — each with its own bounded context, tool surface, and machine…
A developer built FuelLedger AI, an offline-first ledger and AI copilot for independent fuel station operators, particularly in India, that runs Google's open-weight Gemma 3 4B model locally through O…
A developer built QuoteCraft, an offline web app that converts messy client WhatsApp and Instagram messages into itemized quotations, scope-of-work documents, and ready-to-send replies in roughly 15 s…
A developer built HappyGF, an AI-powered accounts payable assistant that extracts vendor, property and amount details from incoming invoice emails, renames PDFs to an accountant's required convention,…
The operator of NxFlowAI, an AI automation agency, outlined a per-sender email parsing approach for converting property portal lead notifications into a uniform record schema before any AI follow-up i…
A developer built QueueInk, an open-source inquiry-desk tool for solo tattoo artists that uses a small open-weight model to extract fields from client DMs and draft replies while ordinary code handles…
Developer Guangyi Zhao tested REA (Reverse Engineer Anything), an MIT-licensed npm package (rea-agents 6.4.0) that exposes 170 reverse-engineering operations as MCP tools so agents like Claude Code, C…
A developer built FlowMedic, an open-source AI failure investigator for freelancers and small agencies maintaining n8n automations. The TypeScript pnpm monorepo ingests n8n failure events, gathers exe…
A developer built LocalDocuShield, an open-source, local-first AI assistant that runs inference on-device with open models so solo medical, legal and accounting professionals can process sensitive cli…
A co-founder of NxFlowAI (Mumbai) outlined a small scheduler pattern for real-estate lead follow-up automation built around explicit stop rules, using a per-lead FollowUp data model, a three-step sequ…
A developer demonstrated with a self-contained SQLite FTS5 example that a RAG pipeline can retrieve the correct document ID while the text passed to the model omits the answer, because a fixed charact…
ApiBench was released as a deterministic, zero-overhead benchmarking framework for HTTP/1.1, HTTP/2, HTTP/3 (QUIC), gRPC, and LLM agent tool-calling workloads. The tool uses a lock-free asynchronous I…
DeepSeek released the V4.1 Flash 1017 update, which improved all 11 of the model's published benchmarks, including Terminal-Bench 4.0 rising from 31.2% to 42.3% and DeepSWE v1.1 reaching 77%, at rough…
A developer at ProLevelMax detailed the cost and mechanics of pulling full Reddit comment threads into an AI agent, including collapsed branches that standard first-request scraping misses. Using the …