PMB - Local Memory for AI Coding Agents

PMB is still very early, but it is already showing strong organic traction with zero marketing spend. The project has grown to 192 GitHub stars and 7,000+ package downloads. Repository analytics from the last 14 days show 2,593 Git clones, 393 unique cloners, 1,459 total views, and 262 unique visitors. This is a stronger signal than passive visibility because developers are not only viewing the project, but actively cloning and testing it. Distribution is also compounding. PMB is available on PyPI as pip install pmb-ai, published in the official MCP registry, and discoverable through the GitHub MCP Registry and related directories. The project also has a modern documentation site and project website, which makes onboarding easier for developers evaluating PMB for real AI coding-agent workflows. Traffic is already coming from GitHub, the documentation site, Hacker News, Google, LinkedIn, Product Hunt, and other external sources. The addressable audience is large and growing: developers using MCP-compatible AI coding agents such as Claude Code, Cursor, Codex, and similar tools who need persistent project memory that is private, portable, and not locked into one vendor.

PMB is for developers who use AI coding agents daily and are tired of re-explaining their project every session. It is especially useful for people working on proprietary or sensitive codebases because all memory stays local on the developer’s machine with no cloud dependency and no API keys. The main users are indie developers, startup engineers, open-source maintainers, and engineering teams that want persistent AI memory while keeping ownership of their project context. As a newly released open-source project, PMB does not yet have notable enterprise customers, but early adoption is visible through stars, downloads, clones, forks, contributors, and external traffic.

Other, Python, MCP (Model Context Protocol), SQLite (events + entity graph), LanceDB (vector store), sentence-transformers (MiniLM embeddings), rank-bm25 (lexical retrieval), FastMCP. Hybrid BM25 + vector + graph retrieval, running fully on-device.

PMB launched recently and is already building meaningful organic traction with no paid promotion: 192 GitHub stars 7,000+ package downloads through PyPI 2,593 Git clones in the last 14 days 393 unique cloners in the last 14 days 1,459 total repository views in the last 14 days 262 unique visitors in the last 14 days Multiple external contributors, forks, and community activity Published on the official MCP registry, with additional discovery through the GitHub MCP Registry and related MCP directories Modern documentation and project website available for onboarding and technical evaluation Repository traffic also shows early external discovery. Referring sources include GitHub, the PMB documentation site, Hacker News, Google, LinkedIn, Product Hunt, and other developer-facing channels. Popular content includes the repository overview, README, pull requests, actions, issues, releases, and traffic pages, which suggests that developers are evaluating the project technically, not just clicking through. Links: GitHub: https://github.com/oleksiijko/pmb PyPI: https://pypi.org/project/pmb-ai/ Docs: https://docs.pmbai.dev/ Website: https://pmbai.dev/