MemoFS Gives AI Coding Agents Persistent Memory Across Sessions and Subagents

·6 views

AI coding agents like Claude Code, Cursor, and OpenAI Codex lose all context at the start of each new session, causing repeated confusion over project rules and dependencies. MemoFS is an open-source, file-based memory runtime designed to solve this by persisting project knowledge across sessions, subagent spawns, and context-window compactions. It supports three integration modes: lifecycle hooks for Claude Code and Codex, a Model Context Protocol server for Cursor and Copilot, and a TypeScript SDK for custom agent frameworks. Developers can initialize MemoFS in any project directory using a single CLI command, which creates structured memory files including a core briefing and long-term notes store. Once configured, MemoFS automatically injects relevant project context into agent prompts at session start and re-injects it whenever the agent compacts its context window.

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)

NET 10 Lets Developers Run C# Files Directly as Shell Scripts

Microsoft's .NET 10 introduces a feature that allows developers to execute a single C# source file as a script without creating a project file or build directory. A developer tested this capability on a Linux container using SDK 10.0.302, finding that after an initial 10.5-second cold start to restore packages and compile, subsequent runs take roughly 0.3 seconds. The feature supports NuGet package references via inline directives and Unix shebang lines, making C# scripts executable directly from the terminal. Build artifacts are stored in a hidden cache folder, keeping the working directory clean like a standard shell script. The developer concluded that for tasks more complex than a simple one-liner, this workflow is now a practical alternative to bash or Python scripting.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Non-Gaussian Distributions Explained: Why Real-World Data Rarely Fits a Bell Curve

Most real-world datasets — including income, stock returns, and website traffic — do not follow the Normal (Gaussian) Distribution, making it essential for data scientists to understand Non-Gaussian alternatives. A Non-Gaussian distribution is simply any probability distribution that deviates from the symmetric, bell-shaped normal curve. Two key concepts help identify such distributions: kurtosis, which measures the likelihood of extreme outliers rather than peak height, and visual tools like histograms and QQ plots, which reveal whether data approximates normality. Distributions with fat tails (leptokurtic) signal higher risk and more outliers, as seen in stock market returns, while platykurtic distributions have thinner tails and fewer extremes. Misidentifying non-normal data as normal can lead to misleading statistical analysis, underscoring why recognising distribution types is a foundational skill in data science.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Experts Warn AI Bubble Could Trigger Worst Financial Crisis in Nearly a Century

Economist Jim Rickards, former CIA and Pentagon adviser, warns the US faces an AI-driven economic crisis potentially more severe than the dot-com crash, the 2008 financial meltdown, and pandemic-era market collapses combined. Veteran investor Jeremy Grantham and former SEC Chairman Gary Gensler have echoed these concerns, with Gensler stating the next financial crisis will originate from AI. The current AI bubble is estimated to be 17 times larger than the dot-com bubble, with companies like OpenAI reportedly spending three dollars for every dollar earned and losing over a billion dollars monthly. Analysts have drawn parallels to the 2008 subprime mortgage crisis, noting that debt used to finance AI data centers is being repackaged into securities and sold to pension funds, mirroring dangerous pre-2008 financial engineering. Additional concerns include circular financing that artificially inflates demand and mounting physical limits on GPU performance and data center energy consumption.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Chinese AI Models Cost Up to 50x Less Than GPT-5.5, But Access Remains a Hurdle

Several Chinese AI models, including DeepSeek, GLM, Kimi, and Qwen, are priced significantly lower than leading Western models such as GPT-5.5, with some costing up to 50 times less per million tokens as of August 2026. Independent benchmarks and public coding challenges suggest these models are competitive in quality for tasks like coding, reasoning, and long-context processing. A major barrier for Western developers is payment access, as official APIs from providers like DeepSeek and Alibaba Cloud often block international credit cards or require Chinese entity verification. API aggregators that accept standard credit cards and offer OpenAI-compatible endpoints have emerged as a practical workaround for developers seeking legal, reliable access. Experts caution users to avoid gray-market resellers, flagging warning signs such as cryptocurrency-only payments, prices far below official rates, and unverifiable service agreements.

0 comments Read more at DEV Community