myc: open-source local memory tool stops AI coding agents from forgetting context

·1 views

A free, MIT-licensed tool called myc (short for mycelium) has been released on GitHub to address a common frustration with AI coding agents like Claude Code and Codex, which lose agreed-upon decisions when their context window is compacted. The tool stores tasks and decision logs locally in a single SQLite file alongside the project, requiring no cloud API or API key. A PreCompact hook writes the session to disk just before compaction and reinjects a recovery packet into the surviving context, preserving earlier decisions. Memory entries are anchored to specific code segments and automatically downweighted if that code is deleted, while decisions extracted from transcripts remain unconfirmed candidates until a human approves them. Currently, myc supports only Bun runtime on macOS and Linux in single-user mode, with no team or server functionality planned yet.

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)

Tutorial: Using Oura Ring Sleep Data and Random Forest to Predict Developer Fatigue

A developer tutorial published on DEV Community demonstrates how to build a fatigue prediction model by combining wearable health data with machine learning. The project uses the Oura Ring API to pull physiological metrics such as heart rate variability, sleep stages, and temperature deviation. These signals are processed with the Polars DataFrame library and fed into a Scikit-learn Random Forest regressor to generate a Cognitive Load Score between 0 and 100. Productivity labels can be sourced manually or synced from GitHub pull request velocity data to train the model. The tutorial aims to help developers move beyond guesswork about burnout by grounding productivity insights in measurable biological data.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Open-Source SOC 2 Evidence Collector Offers Free Audit Prep Tool for Small Teams

Developer Suman Shashikumar has built and released an open-source tool called SOC 2 Evidence Collector, designed to help small engineering teams automate audit evidence gathering without paying for commercial platforms. The lightweight tool connects to AWS and GitHub in read-only mode, checking controls such as IAM MFA status, CloudTrail logging, S3 bucket access settings, and GitHub branch protection rules. Each finding is automatically mapped to the relevant SOC 2 Trust Service Criterion, replacing hours of manual screenshotting and log exporting before an auditor's review. Built on a FastAPI backend with a simple frontend dashboard, the tool is positioned as a free starting point for a team's first audit cycle rather than a replacement for enterprise solutions like Vanta or Drata. Development surfaced real-world API edge cases, including differences between GitHub personal accounts and organizations that can cause silent, misleading responses around branch protection status.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

MyZubster Metaverse Rolls Out Server-Side Safety Tools for Virtual Rooms

MyZubster has detailed a set of server-authoritative safety and moderation features built into its metaverse platform. Virtual rooms follow a strictly controlled lifecycle — from draft to live to ended — with all state transitions validated on the backend to prevent client-side manipulation. Hosts can manage participants, issue expiring single-use private invitations, and control stage access through a moderation interface that avoids exposing internal account identifiers. Room-scoped chat is tied to verified participant sessions, subject to rate limiting and sanitization, and messages expire automatically after 24 hours. The platform currently uses resilient polling for session synchronization, with a dedicated real-time system planned for future integration.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

GEO Experiments: Third-Party Brand Mentions May Outweigh Owned Content in AI Search

Two generative engine optimization (GEO) experiments suggest that mentions of a brand in credible third-party sources may carry more weight than a brand's own website content when AI-driven search tools generate responses. While owned website content remains important as a foundation for clarity and credibility, the research challenges the assumption that optimizing on-site pages alone will reliably boost AI visibility. The findings point to a more source-focused strategy, where marketers identify which trusted publications and resources AI platforms repeatedly cite for relevant topics. Businesses are advised to build legitimate presence in those sources through sustained PR efforts, editorial coverage, and content collaborations rather than relying solely on publishing more pages. A practical starting point is mapping the sources that recur in AI-generated answers for key topics and assessing realistic paths to earning mentions within them.

0 comments Read more at DEV Community