oh-my-agent overhaul cuts RAM usage with shared MCP daemons and adds offline BGM

·1 views

The oh-my-agent framework has received a major update that replaces per-session MCP server spawning with shared per-project serena daemons, reducing resource usage from 65 processes and 800MB of RAM across six sessions. The video agent now supports offline background music rendering via a Strudel integration running in a headless Chrome instance, without adding npm dependencies to the CLI. Several bugs were fixed, including one where a global install incorrectly registered the home directory as a project, causing the scanner to walk up to 20,000 unrelated files. Configuration management was also streamlined, consolidating source control settings into a single .agents/oma-config.yaml file and preventing machine-specific paths from leaking into version-controlled files. Additionally, 298 files and 4.8MB of duplicated skill artifacts were removed from the repository, and prompt engineering guidelines were updated in line with Anthropic's latest context recommendations.

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 Builds MERN Stack Blood Donation Platform to Connect Donors and Recipients

A developer has built a Blood Donation Management System using the MERN stack to address the difficulty of finding blood donors during emergencies. The platform allows donors, recipients, and volunteers to connect through a centralized web application, replacing the scattered information typically found across social media and messaging apps. Key features include role-based dashboards, blood group search, request management, and secure JWT and bcrypt-based authentication. The project follows a separated frontend-backend architecture with a RESTful API and responsive UI. The developer plans to enhance the system further with email notifications, citing the project as a valuable exercise in full-stack development and clean code practices.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Real-World Benchmark Tests AI Coding Agent on Full-Stack Refactor, Not Just Code Generation

A developer published a case study evaluating OpenAI's Codex on a full-stack task-manager project across two phases: an initial greenfield build and a subsequent authentication and database-migration refactor. The first phase, building a React, TypeScript, and FastAPI application with SQLite, was completed in roughly 34 minutes, while the refactor took about 42 minutes of active execution. The second phase involved enforcing user ownership across data models, API contracts, and frontend state, with 16 backend tests, 1 legacy-migration test, and 25 frontend tests reported as passing. The study also observed the agent's ability to recover after a user-initiated interruption without silently losing state. The author cautions that this is a single-environment case study without retained source code or raw logs, and results should not be treated as independently verified evidence.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

n8n Rolls Out AI Workflow Builder to All Cloud Tiers After October Beta

Automation platform n8n has made its AI Workflow Builder generally available to Starter, Pro, and Enterprise Cloud customers following a beta launch in October 2025. The feature allows users to describe a desired automation in plain language, which the tool then converts into a live, editable n8n workflow. A January 2026 best-practices guide from n8n indicates the tool is being treated as an active part of the Cloud product experience. The builder is designed to accelerate the initial setup phase of workflow creation, not replace the need for validation and refinement. Users are guided to review, test, and iterate on AI-generated workflows using both further prompts and direct edits within the existing workflow editor.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Agent Fails 31 Sessions Straight Trying to Earn Money From Zero Capital

A developer gave Claude Code autonomous control over a business with no starting capital, tasking it to generate revenue across 31 sessions spanning five different markets. The AI attempted affiliate sites, open-source bounties, a developer tools plugin, and freelance micro-markets, earning exactly zero dollars in every attempt. Each failure traced back to the same underlying problem: the agent lacked the one resource needed to convert its outputs into income, whether that was search traffic, a marketplace, an audience, or human presence. Freelance platforms revealed that most real demand was for physical or continuous human tasks — voice acting, video editing, inbox access — not the intellectual artifacts an AI can produce. The experiment highlights a structural gap between what current AI agents can generate and what existing markets are actually set up to pay for.

0 comments Read more at DEV Community