Visual Agent Map

English · 繁體中文

Explore big questions with visual AI research in Obsidian

Start with a question. See the pieces. Follow the ideas worth exploring. Build a visual research map, or talk through a new idea in Coffee Tables. Everything you keep is editable Markdown in your vault.

Get started · See the 0.34.0 release · Installation guide

Two ways to explore

• Visual maps. Break a question into topics and research any branch with AI, or check a claim with hypothesis research (Beta), which looks for evidence and stops when it is not enough. Draft a multi-level outline, chat with any single topic to research, expand or synthesize it, pick a note style for each topic type (section structure, voice, format rules and how a note is updated, such as Textbook, or AI suggested for general topics; or write your own in Markdown), merge related topics, research several topics in the background, give a map its own data sources, and mark points on a picture with anchors linked to topics. Manual mapping needs no AI.
  • Talk to your map. Open the AI chat beside the map and say what you want in plain language: build a whole map from one sentence, add, move or rename topics, research or expand a branch, find pictures, place anchors on a selected annotated image, or edit the notes you point it to. Large changes are shown as a plan first, and every action gets a result line.
  • Quick or Standard. Quick answers from the model's own knowledge without the web; Standard searches the web and may add images. Choose per research or expansion.
• Coffee Tables. Bring an unfinished idea to a conversation with simulated AI guests, then revisit the observer's insights.

When a map is ready, save it as one HTML file you can open in any browser, even offline (images that link to the web need a connection), or send to someone else, optionally with AI-translated versions.

Get started

1. In Obsidian Desktop → Settings → Community plugins → Browse, find Visual Agent Map, then install and enable it. Requires Obsidian 1.13.7+ on desktop.
2. VAM opens a blank mind map on first use: start with one sentence, or add topics yourself. For AI features, install and sign in to Codex CLI or Claude Code, then check the CLI in VAM settings.

The interface is available in English, Traditional Chinese and Simplified Chinese. AI answers follow the language you write in, including Japanese, Korean, Spanish and other languages.

Setup details, troubleshooting, and known limitations are in INSTALL.md.

Contributing checks

• npm run verify runs the same checks as CI (lint, format, tests, committed artifacts).
• Run git config core.hooksPath .githooks so it runs automatically before every push.
• Never bypass it with --no-verify.

Privacy

VAM has no telemetry and does not store API keys. AI tasks send relevant content through your local CLI to OpenAI or Anthropic, so your provider allowance or charges may apply. The map AI chat sends the map outline, the titles of your other maps, the cards you select or reference, an annotated image you select (the picture itself), the note open beside the map (in full, shown as a removable chip), notes or text you drop in, and the last few messages; it can edit those notes and create new ones in your vault. Research, expansion and synthesis also read the items in the map's data sources. Saving map AI conversations is off by default; when turned on, they are stored in your vault. VAM runs the Codex or Claude Code program you installed; when you use Standard research or picture search, that program searches the web for you (Quick mode does not). Thumbnails of picture candidates are loaded directly from the websites that host them. A share file is saved in your vault or the place you choose, and VAM never uploads it; whether to publish it is up to you. Images in a share file that link to the web are loaded by the reader's browser from the original site.

Credits and license

The code is licensed under Apache-2.0; the Visual Agent Map name and logo are not, so forks need their own name (see NOTICE).

Report an issue · Changelog · Apache-2.0

Copyright © 2026 Kevin Tsai