Bitweave Promises Sub-Millisecond Hybrid Vector Search Under 1.1 MB RAM

·1 views

A developer has released Bitweave, an open-source hybrid retrieval engine written in Rust with Python bindings, designed for local RAG applications, edge agents, and serverless AI pipelines. The tool aims to address high memory overhead common in standard vector databases, which typically require hundreds of megabytes to gigabytes of RAM. Bitweave uses zero-copy, SIMD-accelerated architecture to handle both categorical filtering and vector search while keeping its active heap footprint under 1.1 MB RSS. Benchmarks shared by the developer show query latency under 1 millisecond, compared to roughly 18 milliseconds and 45 MB RAM usage for an indexed SQLite approach. The full benchmark suite, covering 200,000 records, is publicly available on GitHub for independent verification.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How to reduce Dependabot PR noise with grouping and custom update schedules

GitHub's Dependabot automatically keeps project dependencies up to date, but its default settings can overwhelm repositories with a high volume of pull requests. A Microsoft open source project tackled this problem by adjusting how Dependabot operates. The team applied three key strategies: grouping related dependency updates together, slowing the frequency of routine update checks, and ensuring security fixes still arrive quickly. These configuration changes significantly reduced pull request clutter without sacrificing timely responses to vulnerabilities. GitHub shared the approach on its official blog as a practical guide for teams facing similar challenges.

0 comments Read more at GitHub Blog

ProgrammingHacker News ·

YC-backed Tokenless offers AI API gateway that cuts costs via dynamic model routing

Tokenless, a Y Combinator S26 startup founded by Rohit, Andrew, and Kev, has launched an API gateway that automatically routes AI agent requests between different models on a turn-by-turn basis to reduce costs. The system directs more complex tasks to powerful frontier models while handling simpler ones with cheaper open-source alternatives. The core routing technique, developed by Rohit during his Princeton PhD research, queries multiple models simultaneously and uses their outputs to inform routing decisions. The team claims their router can match the performance of Claude Fable 5 at roughly half the cost, in part by preserving cache state during model switches. Tokenless is now publicly available at usetokenless.com, offering new users $20 in free credits to test the service.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

Step-by-Step Guide: Build a React To-Do App Connected to Supabase

A beginner-friendly tutorial published on DEV Community walks developers through connecting a React application to Supabase, an open-source backend-as-a-service platform. The guide targets those with basic React knowledge and requires no prior Supabase experience to follow along. It covers setting up a Supabase project, creating a PostgreSQL todos table, and integrating the Supabase JavaScript client into a Vite-based React app. The tutorial demonstrates full CRUD functionality — creating, reading, updating, and deleting tasks — using Supabase's auto-generated APIs. Security considerations such as Row Level Security and environment variables for API keys are also briefly addressed.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

intervals-icu Library Offers Typed TypeScript Client for Intervals.icu API

An open-source TypeScript package called intervals-icu provides a structured, typed client for interacting with the Intervals.icu training API. The library, now at version 2.2.1, organizes API operations by resource type—such as athletes, activities, wellness, and events—rather than exposing a single large interface. It supports both API key and OAuth authentication, handles transient failure retries automatically, and requires Node.js 18 or newer with ESM module support. Developers can install the package via npm, initialize a client with credentials stored as environment variables, and query athlete data with minimal boilerplate code. The repository is publicly available under the MIT license, and the tutorial focuses on establishing a maintainable client foundation rather than building a full training application.

0 comments Read more at DEV Community