OKF Agent Memory gives coding agents persistent, git-native knowledge across sessions

·4 views

OKF Agent Memory, an open-source tool released on September 5, 2026, aims to solve the problem of coding agents losing architectural context between sessions. The project stores an agent's knowledge as plain Markdown files with YAML frontmatter inside a repository's knowledge/ directory, making memory reviewable and version-controlled like any other source file. It ships as a single dependency-free Go binary featuring BM25 search with sub-300-microsecond lookups, requiring no vector database, embedding API, or external accounts. The tool exposes a stdio MCP server compatible with agents such as Claude Code, Cursor, and Codex, allowing them to query stored concepts directly. Version 0.4.4, released September 27, 2026, has already accumulated over 700 GitHub stars in under four weeks and topped weekly trending charts for AI agent repositories.

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)

Study of 8,022 MLB Pitch Pairs Finds Whiff Rate More Repeatable Than Run Value

A data analyst examined Baseball Savant pitch statistics from 2017 to 2025 to determine how consistently five pitch-level metrics carry over from one MLB season to the next. Using 8,022 pitcher-and-pitch-type pairs with at least 100 pitches in both seasons, the study measured year-over-year correlations for usage rate, whiff rate, xwOBA allowed, hard-hit rate, and run value per 100 pitches. Whiff rate proved the most stable quality metric, with correlations ranging from 0.53 to 0.74 depending on sample size, while run value was the least repeatable, peaking at just 0.35 even for pitches thrown 800 or more times. All five metrics showed stronger correlations as pitch counts increased, consistent with the expectation that larger samples reduce statistical noise. The findings suggest that a single season of run value is a weak predictor of future performance, whereas whiff rate offers a more reliable signal of a pitch's true quality.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why a Locally Working Backend Is Not Automatically Production-Ready

A backend application that runs flawlessly on a developer's local machine may still fail in production, where the environment is no longer controlled or predictable. Local setups typically have pre-configured variables, stable databases, and forgiving conditions that production systems do not guarantee. Critical gaps include missing environment variable validation, risky database migrations that can break live data, and process managers that report a server as running even when it cannot serve requests. Production systems also require meaningful health check endpoints and structured logging, since console logs become nearly useless when errors occur with no one watching the terminal. Addressing these gaps before deployment — through incremental migrations, startup validation, and proper observability — is what separates a working prototype from a reliable production service.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Cloudflare Launches cf CLI and Open-Sources Forge SDK Generator for AI-Ready Workflows

Cloudflare has released cf, a command-line tool that covers its entire API surface and supports TypeScript configuration-as-code, making it usable both by human operators and AI agents. Alongside the CLI, the company open-sourced Forge, its internal SDK generator that automatically produces typed TypeScript code from OpenAPI specifications. Because the CLI is generated rather than hand-written, it stays in sync with API changes without requiring manual updates to documentation or flag definitions. The design reflects a growing trend in infrastructure tooling, where CLIs are being built with AI agents as first-class users alongside humans. Cloudflare's approach prioritizes complete API coverage and programmatic usability over a curated, beginner-friendly command set.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

HexStrike AI Brings 150+ Offensive Security Tools to AI Agents, Raising Sandbox Concerns

HexStrike AI is an open-source MCP server that connects large language models like Claude and GPT to over 150 offensive security tools, including exploitation frameworks, vulnerability scanners, and password attack utilities. The project has gained significant traction on GitHub, accumulating more than 12,000 stars and ranking among the top trending Python repositories. By acting as a bridge between AI agents and these tools, the server enables autonomous penetration testing but introduces serious questions about containment and misuse. Key architectural challenges include preventing tools from attacking infrastructure beyond their intended targets, with proposed safeguards ranging from per-invocation Docker containers to proxy-based network allowlists. The lack of mandatory human approval in the MCP protocol further complicates oversight, prompting debate around permission models and audit trails for high-risk tool invocations.

0 comments Read more at DEV Community