Every challenge multi-agent users face, solved by design.
Agents don't know what you know
Shared Memory
Never remind agents of any context. They read it from the graph automatically.
4-10 terminals is overwhelming
Spatial Navigation
Jump back in and it's exactly how you left it. Spatial memory does the recall for you.
Manual agent coordination
Recursive Decomposition
AI breaks big tasks into subgraphs and spawns parallel subagents on the dependency graph.
Long sessions rot context
Infinite Context, Zero Rot
The graph grows indefinitely. Agents get short, focused sessions with only what's relevant.
Complex tasks make agents unpredictable. Simple, isolated tasks make them reliable. When agents are reliable, you treat their outputs as building blocks — assembling massive projects fractally from tiny tasks.
See it in action
“
Feels like alien technology in a good way... I was hooked
— Lochlan Hill, AI Lead @ Planet Price
Frequently Asked Questions
Will this have access to my data?
Voicetree is local-first: your files stay only on your machine as plain Markdown. When you use AI features, content is sent to third-party APIs (Google, Soniox) for processing — these providers do not store your data for training. We collect anonymous usage telemetry. See our privacy policy for details.
Is this an Obsidian plugin?
It started as one (agent-canvas on GitHub), but outgrew the plugin ecosystem. Voicetree is now a standalone Electron app that opens your existing Obsidian vaults. Your files stay as plain Markdown.
Can I use it without AI features?
Yes. AI is completely optional and off by default. Use Voicetree as a pure spatial markdown editor — AI won't see or modify your notes unless you explicitly launch an agent.
Can I keep my current agent workflows?
Yes. Voicetree runs the same agents you already have installed on your computer in embedded terminals. You can use them exactly as you're already using them.
