Filamental AI

Click the sparkle button in the top bar, pick your provider and paste its key, and Filamental AI reads, builds and edits your open project from a chat.

Where: Top bar › Filamental AI (the sparkle icon) · also Connect an AI in the exploration window

What it is

Filamental AI is a chat window that opens beside Filamental, with live read and write access to your open project and nothing else to install. It's the connection rather than a model of our own: every reply comes from the model you pick, named in the dropdown at the top of the chat.

Your key is saved on this computer, in Filamental's own settings. It and your conversations go only to your provider, so Filamental never sees them.

How to use it

1. Click the sparkle button in the top bar. The Filamental AI window opens beside your project.
2. Choose your provider from the list.
3. Paste your API key and click Connect. If the provider turns the key down, its own message tells you why.
4. Pick the models you want under Models (pick any you'd like to add), then click Add selected models.
5. Ask away in the Ask Filamental AI… box, and press Enter to send.

Have a normal conversation, then say "build this out in Filamental", and it builds it: nodes with notes and properties, the connectors between them, and new classifications and connector types with their colours and weights. It can also find the shortest route between two nodes, the one the Pathfinder lights, and look things up in this help. Each step shows in the chat as a line starting with →.

The providers in the list: Claude (Anthropic), Gemini (Google), OpenAI, DeepSeek, Kimi (Moonshot), Qwen (Alibaba Model Studio), Grok (xAI), Mistral, GLM (Zhipu / Z.AI), MiniMax, Groq, OpenRouter, Together AI, Fireworks AI, SiliconFlow, Cohere and Venice AI. Kimi, Qwen and MiniMax each have a second, China entry, so pick the one your key came from. Ollama (on this computer) and LM Studio (on this computer) are in the same list, covered in AI on Your Own Computer.

If yours isn't there, click Don't see your provider? and tell us. It's usually a small job, so do ask.

Several models: + Add another model, in the dropdown at the top of the chat, connects another: a second opinion, or a cheaper model for quick questions. Disconnect removes the one you're on.

Attachments: the + beside the message box attaches a PDF or a JPG, PNG, WebP or GIF image (up to 15 MB, sent to the model as it is), or a Word or text file, read as text and cut at 20,000 characters, with a note saying so.

Undoing a reply

When a reply changes your project, Undo these changes appears under it and puts back everything that reply did, newest first, as it was before the reply, so your own edits since to the same things go back too. Anything it can't put back, such as a node deleted since, is counted on the button. The main window's undo doesn't reach these changes, see Undo and Redo.

Tips

• Switching model starts a fresh chat, as does closing the window. The project keeps everything built.
• A long build may pause itself after a long run of steps, to avoid a runaway loop on your key. Ask it to continue.
• It works on the project that's open and can't create a project folder, so open or create the project first.
• On the free plan it builds within the project's node limit, like you do. See Plans: Free and Pro.
• An AI can write a property your project doesn't define. It shows on that node with an ×, see Properties.

How this helps you

Use it to build in conversation without leaving Filamental, to question a project ("what's missing?", "how is this connected to that?"), or to read a document you attach and add what matters.

Related

• AI on Your Own Computer: Filamental AI on a local model.
• Explore a Folder or Document with AI: point it at a folder of documents.
• Connect an AI App: the same reach from Claude Desktop or another app's window.
• What We Tell Your AI: the briefing it starts every chat with.