Introduction to Agent Bayes

• Introduction to Agent Bayes
  • 👋 Welcome to Agent Bayes!
    • This mindmap introduces what Agent Bayes is and what you can do with it. Please invest two minutes here and the rest of the product will make sense.
    • The left sidebar has a sample project, My Project, selected, with its mindmaps listed underneath.
    • You are looking at one of those mindmaps. It is the same kind of map you will build for your own research, out of your own papers.
  • 📚 The agent works from your knowledge base, and nothing else
    • A knowledge base is your indexed library. PDFs you upload, or items you sync from Zotero, get indexed page by page into a corpus the agent can search.
    • That corpus is the only place the agent retrieves from. It does not search the web, so every citation made traces back to a file you put in a knowledge base of your own.
    • A project serves as the scope. It holds the mindmaps, and the agent can retrieve only from knowledge bases that are attached to the project you are working in.
  • 🗺️ This mindmap is your workspace
    • A mindmap is a tree of nodes. Each node holds one piece of your argument, and the tree holds the pieces organized hierarchically.
    • It persists. Come back in three weeks and the structure is where you left it, with your reasoning still legible instead of in scattered notes or fourteen Zotero tabs.
    • Everything here is editable. Write your own nodes, rewrite the agent's output, move nodes around the map, delete what you don't need. Undo and redo cover every change, yours and the agent's.
    • There are a lot of ways to work with a mindmap, but organizing sibling nodes at the same "level of abstraction" works well, because missing aspects immediately surface.
    • This map is a sample. Read it, cut it apart, rename it, or delete it once it has served its purpose.
  • 🚀 A basic workflow to try
    • 1️⃣ Add papers into a knowledge base. Upload PDFs directly, or sync items from your library with the Zotero plugin.
    • 2️⃣ Create a mindmap and ask for something broad. For example: build an overview of the leading explanations for X, with their main proponents and the evidence behind each.
    • 3️⃣ The conversations with the agent sit in the right sidebar. The tabs beside it hold your citations, and other tools such as semantic search across your entire corpus and your labeled passages.
  • 🤖 What the agent does with your papers
    • It plans the work, retrieves passages from your knowledge bases, and writes synthesized claims or lists evidence straight onto your mindmap as nodes, backed by citations you can open.
    • It writes continuous prose, not only bullet points. Claims, the sentences that connect them, and the summaries that hold a section together.
    • When your sources disagree, it opens a sibling node for each position rather than averaging them into one flat sentence. You evaluate them separately.
    • The agent drafts. You decide what survives, what gets rewritten, and what the argument actually is.
  • 🔖 Citations are the point
    • A node with citations is a claim backed by your sources. The content says something, the citations show where it came from.
    • Click a citation to open the PDF at the passage behind it, on the right page. Checking a claim takes one click, not a trip back through the folder you saved it in.
  • 🎯 Telling the agent what you want
    • The mindmap agent has the mindmap as context. It doesn't load it fully because it can grow huge. It is able to search within the mindmap, or load parts of it on demand.
    • Selecting a node on the canvas hints to the agent which part of the map you mean, and it can load that subtree if it makes sense. Pinning a node is the stronger move: the node and its subtree will be included with the message you typed.
    • Structure is something you can ask for. Rearrange these children chronologically. Merge these two siblings. Move the methodology nodes under a new parent.
    • Follow-ups beat restarts. Push on one branch until it says what you mean, then move to the next one.
    • When a map is worth showing, you can choose to publish a current snapshot of it and make it publicly visible.