Agentic memory is arguably the most critical frontier in AI right now. As autonomous agents move from simple chat interfaces to long-running background processes, they need a place to store context, facts, and user preferences.
Currently, the industry standard is moving towards solutions like Mem0.ai. They have built phenomenal infrastructure. But they are operating on a paradigm designed for enterprise SaaS, not for true human-machine symbiosis.
With Matryca Plumber, an autonomous background AI daemon built on top of Logseq OG, we are proposing a radically different approach: the Local-First Document Plane.
Here is why redefining agentic memory as a shared physical space between humans and AI is the only way to achieve true Sovereign AI—and why this architecture fundamentally outperforms the status quo.
🏛️ The Paradigm Clash: Two Visions of Memory
To understand the difference, we have to look at how these systems handle a simple piece of information. Imagine you tell your AI: “I am working on the Acme Billing Module, and my preferred stack is TypeScript.”
The Industry Standard (Mem0.ai): The “Black Box” Silo
Mem0 and similar frameworks treat memory as a Data-Native abstraction.
- Extraction: An LLM extracts the atomic fact from your prompt.
- Storage: The fact is routed to a hidden, multi-store database (a Vector DB for semantics, a Graph DB for relations, a Key-Value store).
- Retrieval: The next time you mention “Acme”, the system silently injects these facts into the agent’s system prompt.
The result? The AI has a memory, but you are locked out of it. It lives in a proprietary database or a cloud vector store. You cannot easily read it, format it, or link it to your daily notes. It is a memory for the machine.
The Sovereign Standard (Matryca Plumber): The Shared State
Matryca Plumber treats your local file system as the ultimate System of Record.
- Navigation: The daemon searches your actual Logseq knowledge graph (local .md files) for the Acme.md page or today’s Journal.
- Surgical Mutation: It directly modifies the AST (Abstract Syntax Tree) of your Markdown, injecting a clean, structured block:
– Project: Acme Billing Module
tech-stack:: [[TypeScript]]
status:: #active
- Safety First: If you happen to be typing on that exact file at the same millisecond, Matryca’s Ironclad OCC (Optimistic Concurrency Control) detects the file timestamp (mtime) drift, aborts the AI’s write, and re-merges the data so you never lose a single keystroke.
The result? Complete transparency. The AI’s thoughts are written in your files. You can read them, tag them, or edit them from your phone. Human and machine share the exact same cognitive workspace.
🚀 The Unfair Advantages of Matryca Plumber
This “Shared State” architecture provides massive technical and economic advantages over both cloud-based memory silos and traditional flat-file editors like Obsidian.
1. Block-Level Granularity (The Token Economics Triumph)
Most Markdown-based AI tools (like Obsidian plugins) treat files as flat text blobs. If an agent wants to update a project status, it must load the entire 2,000-word document into its Context Window, ask the LLM to rewrite it, and overwrite the file. This is slow, expensive, and dangerous.
Because Matryca Plumber is built on Logseq’s outliner model, it operates on Blocks. Every bullet point has a unique identifier (id:: 64a…).
- The agent can fetch, read, and mutate a single 20-word block instead of a 2,000-word page.
- This surgical precision consumes a fraction of the compute costs (and LLM tokens).
- It leaves the rest of the agent’s context window entirely free for deep reasoning.
2. Absolute Data Sovereignty
There is no third-party database. There is no cloud API holding your preferences hostage. If you turn off the AI, your memory doesn’t disappear into a proprietary dashboard—it remains perfectly structured in your local Markdown files, forever accessible via Logseq or any text editor.
3. The Foundation for “Matryca Brain”
By keeping the files as the absolute System of Record, we are paving the way for the v2.0 Shadow DB—a lightning-fast, daemon-owned SQLite read-cache that will give the AI instant semantic retrieval speeds (like Mem0), while keeping the actual data safely anchored in your local folders. This shared engine will eventually power higher-level user interfaces, like the upcoming Matryca Brain.
🤝 Join the Revolution: How You Can Help
Matryca Plumber is not just a tool; it is a movement towards truly open, local-first, and privacy-respecting AI collaboration. Building an enterprise-grade agentic OS takes a village.
Here is how you can help us push this vision forward:
💻 For Developers: Contribute Code
Our codebase is rigorous (Strict Mypy, Ruff, 700+ passing tests, zero-regression architecture), but we have plenty of accessible entry points.
- Tackle “Good First Issues”: Check our GitHub repository. We have clearly labeled tech-debt and feature issues perfect for your first PR.
- Expand the Agentic UX: Help us build dynamic MCP tool filtering, refine our Markdown AST parser, or write new memory modules.
- Review & Test: Clone the repo, run make check, and help us harden the daemon against edge cases.
❤️ For Believers: Sponsor the Project
If you are not a developer but believe in a future where AI works with your local data rather than hoarding it in the cloud, you can fuel this project directly.
- GitHub Sponsors: Your financial support allows me to dedicate more hours to complex architectural challenges (like the v2.0 Shadow DB) instead of standard client work. Every tier helps pay for local compute, API testing, and coffee. Go here for sponsor this project
Let’s stop building brains for AI, and start building a shared brain for humans and machines.
Check out the repository at https://github.com/MarcoPorcellato/matryca-plumber , read the ROADMAP.md, and join the local-first movement today.