Egregor Tool Deploys Multiple AI Models Locally to Reduce Hallucinations and Guard Privacy

·4 views

A developer named Vladislav Shter has released Egregor v1.1, an open-source desktop application that routes user queries through a council of specialized AI models rather than a single chatbot. The tool runs entirely on the user's local machine, sending only fragmented context to each AI model to protect intellectual property and data sovereignty. To counter groupthink, Egregor uses a blind first round, a rotating devil's advocate role, and a red team final round where models challenge the emerging consensus. A confidence scoring system rates each final output on a scale of one to five, helping users gauge reliability. Pricing ranges from free for single-model queries to around two dollars for full codebase audits, with source code available on GitHub.

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)

Developer Rebuilds 1991 DOS Maze Algorithm in the Browser After 35 Years

A software developer has revived a maze-generation program originally written in Turbo Pascal on a DOS machine in 1991, porting it to the browser using TypeScript. The algorithm was not intentionally designed as a maze generator — it emerged from experiments with recursive tree-drawing routines, where branches arranged at 45-degree angles naturally produced cycle-free paths. The developer notes that the approach is functionally equivalent to randomized depth-first search, a technique that gained wider recognition through Jamis Buck's 2011 blog series on maze algorithms, roughly two decades after the original code was written. No invention claim is made; rather, the author describes independently arriving at the same method through visual and geometric thinking rather than formal algorithm study. Notable design choices from the original code — including a precomputed permutation table and using pixel color as the sole occupancy map — have been preserved in the modern rewrite.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Entire Launches Distributed Git Mirror Network Built for AI Agent Workloads

A startup called Entire has released a preview of a distributed Git network designed to handle the high-frequency clone and push operations generated by large fleets of AI coding agents. The system places regional mirrors in the US, EU, and Australia in front of an existing GitHub repository, absorbing read and write traffic without touching the origin or its existing CI and access controls. Initial benchmarks on a single repository recorded roughly 570,000 shallow clones per hour, 586 pushes per second, and a mixed clone-plus-push throughput of around 470 operations per second at 50–60 ms median latency. The product is currently available under a waitlist, and Entire has said it plans to open-source the underlying Git backend in the coming months. Future milestones include native repository hosting and a fully decentralized network to support enterprises with data residency and regulatory requirements.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Build a Real-Time Dashboard Using FastAPI, WebSockets, and Postgres

Traditional polling-based dashboards send repeated server requests every few seconds, wasting CPU, bandwidth, and delivering stale data to users. WebSockets offer a more efficient alternative by letting the server push updates to clients instantly over a single persistent connection. Using FastAPI's native WebSocket support alongside PostgreSQL's built-in LISTEN/NOTIFY feature and the asyncpg library, developers can build a lightweight real-time dashboard with minimal infrastructure. For single-instance deployments, this setup requires no external message brokers and is straightforward to maintain. When scaling horizontally across multiple backend pods, adding Redis pub/sub ensures all instances receive and broadcast every database event to their connected clients.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How IVF and HNSW Power Fast Vector Search in Modern AI Systems

Vector search underpins semantic AI applications like RAG pipelines, recommendation engines, and image search by finding the closest matching vectors among millions. Because high-dimensional embeddings suffer from the curse of dimensionality, exact nearest-neighbor methods break down and approximate nearest neighbor (ANN) search is used instead. ANN algorithms trade a correctness guarantee for speed, measuring quality through recall — the fraction of true top-k neighbors actually returned. Two algorithms, Inverted File Index (IVF) and Hierarchical Navigable Small World (HNSW), handle most of the work in popular libraries like pgvector, Qdrant, and FAISS. Understanding how each works under the hood helps engineers choose the right approach and tune the speed-versus-recall tradeoff for their use case.

0 comments Read more at DEV Community