Filamental has no AI of its own to sell you. It works with it on or off and it costs exactly the same either way. If you do want it, five routes are available, and you bring the model you already trust.
There’s no assistant in the corner of the window. Nothing suggests, nothing completes a sentence you were still thinking about, and nothing is quietly reading over your shoulder. Open Filamental with no key, no connection and no plugin and you have everything you need.
In a different version of this company you’d be meeting Philomena, the Filamental Agent, Filly to her friends, waving at you from the corner of the window. We didn’t build her.
You’re already paying for a very good model, you have already learned how to talk to it, and you have already tailored it to the way you work.
No branded agent
We could ship a chatbot like everyone else, but we both know what that is: somebody else’s model, marked up and sold back to you as a unique selling point, with the U doing a great deal of heavy lifting.
Model agnostic, and staying that way
Frontier or open weights, cloud or local. We would rather you don’t feel tied to any one model.
Having options when it comes to AI is, and will be, important.
We can’t upsell and meter what we don’t offer.
Talk a subject through in the chat window you already use, then ask it to project it into Filamental. A minute later the structure’s standing there, named and connected, and there’s no blank canvas to face.
Point Filamental at a folder of documents (or just one long one) and your AI reads them and builds a first draft of the structure. You approve the vocabulary and see a token estimate before it starts, and nothing in the folder changes.
Ask it to finish the parts you didn’t, such as notes on nodes you moved on from or the connector type you kept meaning to define. It works inside the space rather than writing a document about it.
Hand it the whole project and it gets your reasoning, not just your files: what depends on what, which way round, and how strongly. Ask how two things connect and it reads back the chain, the same route Pathfinder lights up on your screen.
What an AI hands back is usually a document, and documents flatten the thinking. A space keeps the categories and the dependencies, and since it’s a folder of Markdown, whoever you send it to can point their own AI at it and read it straight back.
None of that needs a translation layer, because there's nothing to translate. A node is a Markdown file with its relationships written down in the front matter, which is a format every current model has read a great deal of.
You didn't do anything extra to get that. It's incidental organisation, and it's what's left behind by working the way you wanted to.
You’ll recognise yours on sight. If none of these words mean anything, this section just isn’t for you, and that’s fine.
MCP
Claude Desktop, Claude Code, Codex, Cursor, Gemini CLI, VS Code, Zed, LM Studio, and anything else that speaks it. A live connection to the space you have open.
Your own API key. Any model.
Paste a key from whoever you already pay, or run a model on your own computer through Ollama or LM Studio. It always says which model you’re talking to, and we never see the key or the conversation.
Point it at the folder
Anything that reads files can read a space, because a space is files. The second brain route, for people who already built one.
Export for AI
The whole structure as one Markdown file, with a short summary at the top so the model knows the shape before it reads a note. Attach it to any chat box, including the one your employer gives you.
Mermaid
Build it in conversation, paste the code, and it becomes a real space with types you assign as it imports. If you know the word, you already know what this does.
Each of these has a page of its own: Claude Desktop, Claude Code, Codex, Cursor, Gemini CLI, VS Code, Zed, LM Studio, Obsidian, Mermaid and plain Markdown files, or all of them together.
Every one of these is free, on every plan, permanently. The full mechanics live in AI Connections & Prompting, where they belong.
Filamental’s own documentation is a Filamental space, one node per topic, built and kept up to date using the routes above. We reached for it because writing documentation any other way had become the slower option.
That one space generates the Help section on this site and the help inside the app, so they can’t drift apart. Go and read it: you’re looking at the output.
Every route needs the model to understand a Filamental file first. That briefing is two plain text files, published here word for word as they’re sent: the skill file and the file format reference.
If software is going to brief an AI on your behalf, you should be able to read exactly what it says.
Download it, never press the AI button, and nothing you’re using is diminished. It’ll be there if you change your mind, and it’ll still be free when you do.