Point an AI that can read your files, such as Claude Code or Cowork, at your project folder and give it the two briefing files, and it reads your notes as what they are, plain Markdown on disk, and writes new nodes straight back into the folder.
Where: your project folder, with the files from Settings › AI › Files for other AIs
What it is
Every Filamental project is a folder of Markdown files, one per node, with its classification, properties and connectors written at the top. An AI with access to your files can work on that folder directly, without connecting to Filamental at all. Teach it the format with Files for Other AIs and it can read your whole structure, reason across it, and write properly formed new nodes.
This is the route that turns Filamental into a working second brain rather than a graph you consult. The AI isn't visiting your structure through a window. It's working in the folder alongside everything else you keep there.
How to use it
- Download both files from Settings › AI › Files for other AIs, the Project Assistant and the Format Reference, and install them in your AI as skills or instructions.
- Give the AI the project folder, the way that AI takes a folder.
- Ask for what you want: a gap analysis, new nodes, a briefing, or a whole new project from scratch. With the Project Assistant loaded, it reads your classifications first, gives you a short orientation and asks what you'd like to do.
While the project is open in Filamental, files the AI writes into the folder appear as nodes by themselves.
Folder access or MCP?
Why you'd choose the folder:
- It sees everything in the folder, not just the graph: subfolders, attachments, PDFs, spreadsheets, whatever else lives alongside your nodes.
- It can work across several project folders at once, and across files that have nothing to do with Filamental.
- It doesn't need Filamental, or the index, so a folder someone sends you is enough.
Why you'd choose MCP (Connect an AI App):
- MCP hands the AI a resolved graph it can query: search every node, follow a connector, pull out a neighbourhood, with arrow directions already worked out. Over the folder the AI reads raw files and reconstructs all that itself, which is slower and easier to get wrong once a project grows.
- Writes through MCP are checked. Writing files directly depends on the AI following the Format Reference, so scan what it produces.
They're the same idea at different depths, and neither replaces the other.
Tips
- Check the links it writes. An AI writing files by hand can occasionally produce a malformed UUID, or point a connector at one that matches no node. Every relationship's
target should appear as the id of another node.
- Point it at the right folder. Build and reason against the project that's still moving. A finished project archived into Obsidian is reference material, which an AI reads just as well, but nothing is restructuring it any more. See Working with Obsidian.
- Key values are plain text in the files. A Key property is hidden in Filamental, but anything reading the folder reads it. See Properties.
How this helps you
Use it when you want an AI to hold your whole working context rather than one structure: the project, the documents behind it and the folder of research beside it, all in one session.
Related
