SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

Developer builds open-source multi-agent coding terminal with 3D office visualization

0
·1 views

A developer named Dominque Church has released Gnosis, a free, open-source, local-first terminal coding agent built with a TypeScript backend and a React plus Three.js 3D office interface. The tool visualizes AI agents as entities moving through five distinct zones — including planning, coding, and testing — so users can see what each agent is doing in real time. Gnosis supports any OpenRouter language model and separates conversation state from the model provider, allowing mid-session model switching without losing context. A prompt-caching optimization reportedly cut one request's cost from $0.0252 to $0.0021, a roughly 12-fold reduction. The project is MIT licensed, ships with 104 automated test suites, and is installable via npm.

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)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

How to Build a Fully Offline AI Finance Assistant Using LiteRT and Gemma 4

A developer guide published on DEV Community outlines how to build a privacy-first personal finance assistant that runs entirely within a web browser without any internet connection. The application uses Google's LiteRT LM Web API and the Gemma 4 E2B small language model, caching roughly 2GB of model weights in the browser's Cache Storage for offline use. Transaction data is stored locally via IndexedDB, ensuring sensitive financial information never leaves the user's device. To reduce AI hallucinations on numerical data, the system pre-computes expense totals by day, month, and category before passing them to the model as verified ground truth. The tech stack includes Angular 22, TailwindCSS, and Node 24, with conversation memory managed carefully to stay within on-device token limits.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

EvoGuard Aims to Add Context-Aware Safety Layer Between AI Code and Production

A new open-source platform called EvoGuard is being developed to sit between AI-assisted code generation tools and a repository's production branch. The platform evaluates pull requests not just on their code diff but against the broader repository context, including past rollbacks, incidents, architectural conventions, and dependency history. Its core premise is that AI coding agents can produce syntactically valid, test-passing code that still violates undocumented but critical codebase conventions. EvoGuard aims to convert fragmented repository history — spread across Git logs, CI systems, and incident reports — into structured 'repository memory' that informs future merge decisions. The project is currently in development and is available on GitHub under the handle modarresi1913/Evoguard.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Social Posting Automation Silent for 14 Days While Monitoring Falsely Showed All Clear

A ¥1.2M/month content automation system stopped publishing to Instagram and TikTok on July 28–29, but the failure went undetected for two weeks because social posting lanes were never included in the existing monitoring script. The oversight was not a bug but a design gap — the watchdog tool was intentionally built to cover only article-related pipelines, leaving social channels completely unobserved. Compounding the problem, posting scripts consistently exited with code 0 even when no content was actually published, causing the system to register eight out of twelve failed runs as healthy. The operator discovered the outage only by chance while reviewing logs for an unrelated issue. A revised monitoring approach was subsequently implemented, shifting the health check from 'did the script complete' to 'does a success marker appear in today's log,' better aligning technical signals with real business output.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Benchmark: Web3-Focused Scanner Drainscan Outperforms Gitleaks and TruffleHog on Crypto Key Detection

A benchmark test across 500 web3 repositories found that generic secret scanners gitleaks and trufflehog missed approximately 73% of web3-specific private key leaks, while the newer tool drainscan detected significantly more with a 4% false-negative rate. Drainscan identified 487 web3-specific true positives compared to 89 and 112 for gitleaks and trufflehog respectively, and achieved a 3% false-positive rate versus 34–41% for its rivals. Key blind spots in the generic tools included Phantom wallet JSON exports, BIP-39 mnemonic checksum validation, and Solana base58 seeds, all of which drainscan flagged using web3-aware semantic analysis. The benchmark was conducted using default and aggressive configurations with manual verification of findings across Solana, Ethereum, and DeFi repositories. Drainscan is available as a free pip-installable tool, with a one-time $99 Pro license unlocking git history auditing and SARIF reporting.

0 comments Read more at DEV Community