Bio-inspired PKM · Now in beta

Your brain's
operating system

A living knowledge graph that learns from how you think. Connect Obsidian, Claude, and your own notes into one neural network that remembers - and forgets - like you do.

See the features
CYTON|My AI brainclaude · 24 msgs7 reviews
live
Memory
History
Habits
Health
Memory Layers
Sensory38
Working61
Episodic44
Semantic27
Procedural15
Consolidation58%
Emotional wt.22%
Vault Stats
Neurons342
Synapses891
Inbox12
Reviews7
Stageadolescent
Neuron Limit342 / 500
startup strategysecond brainknowledge graph
● concept extracted: "distributed cognition" · synapse strengthened: startup-strategy → knowledge-graph · REM suggestion queued
Connections
REM
Prune
Active Connections
Obsidian
vault connected
Claude Code · S1
active · 24 msgs
MCP Server
127.0.0.1:8765
20 tools · HTTP
Why Cyton

Other tools hand Claude your text.
Cyton hands it your judgment.

Search plugs your notes into Claude as a flat pile of text - a half-finished thought ranks the same as the idea you've sharpened for months. Cyton scores every memory the way your brain does: by how reliable it is, how much you value it, and how central it is to your thinking. So Claude leads with what you actually trust and return to, not the closest keyword match.

It answers from your conclusions, not your scratchpad.

Every note carries a reliability tier and a salience score. A consolidated idea outranks a raw capture; the thoughts you keep returning to outrank the ones you jotted once. Claude leads with your settled thinking, not the first keyword match.

It knows which of your ideas are load-bearing.

Cyton tracks your hubs - the concepts everything else connects to. Ask something broad and Claude anchors on your core ideas first, then follows the strongest links outward, the way your own recall actually moves.

It resurfaces what you'd have forgotten.

Every memory sits on a forgetting curve. When a note you haven't opened in weeks turns relevant again, Cyton flags it and feeds it in - so Claude reminds you of your own past thinking at the moment it counts.

It compounds every time you think.

Retrieval is static - it finds similar text and stops. Cyton rewires around you 24/7: links strengthen when ideas come up together, fade when they don't. Every session starts from a sharper map of your mind than the last.

Notes as flat text
Notes ranked by keyword similarity alone
A throwaway line weighs the same as your best work
No signal for what you value or rely on
Context rebuilt from zero every session
Notes through Cyton
Ranked by reliability, salience, and centrality
Your most-used, most-refined ideas lead the answer
Fading-but-relevant memories resurfaced automatically
The graph rewires as you think - context compounds
Neuroscience-backed

Memory isn't flat.
Your PKM shouldn't be either.

Cyton models the same five memory systems neuroscientists have identified in the brain. Every note you take is classified, decays realistically, and consolidates over time.

Sensory

Raw input, short-lived. Captured automatically from Obsidian edits.

Working

Active context window. What your brain is focused on right now.

Episodic

Events and experiences. Specific moments that carry temporal context.

Semantic

Abstract knowledge. Concepts distilled from many episodes.

Procedural

Habits and patterns. Learned sequences extracted from your behavior.

Cyton - Graph View
342 nodes
knowledge graphstartup strategyLLM agentssecond brain
Search nodes...
Sensory
Working
Episodic
Semantic
Procedural
Knowledge graph

Your notes.
Connected automatically.

Every conversation you have with Claude Code shapes your graph. When topics come up together in a session, Cyton strengthens the edge between them. Topics you stop discussing gradually decay. No manual linking. No tagging required - your thinking does the work.

Conversation frequency between topics determines edge weight
Hub nodes surface your most-discussed, most-connected ideas
Edges decay when topics fall out of your conversations
Spaced repetition queue resurfaces what your graph flags as fading
Graph enrichment

Your notes,
enriched automatically.

On the Free plan, Cyton enriches your graph automatically - assigning tiers, extracting concepts, and linking ideas. To do it, the opening of each note (around the first 600 characters) is sent to Claude Haiku through Cyton's own account. No setup, no API key needed.

On Pro, that same enrichment routes through your own Claude Code instead. The note text goes to Anthropic under your account - Cyton's servers and API key are never in the loop. (See exactly what's sent below.)

Free - through Cyton's Claude account, automatic
Pro - through your own Claude, not Cyton's
terminal - claude code
$
Data & AI transparency

Exactly what Cyton sends to AI.
And what never leaves your Mac.

Cyton keeps and searches everything on your machine. To enrich your notes and grow your graph, it works with Claude. Here is exactly what gets sent to AI, what never leaves your Mac, and what AI writes back - no fine print.

Stays on your Mac
Your knowledge graph
Nodes, links, and connection weights live in a local database on your Mac.
Every memory score
Tiers, salience, centrality, and the forgetting curve are all computed on-device.
Semantic search
Your notes are turned into vectors on your own machine and matched locally. The text is never uploaded to be searched.
Your Obsidian files
Read in place. Cyton never copies, moves, or uploads your vault.
Usage counts and limits
The free-tier daily cap is tracked in local storage, with no server call.
Sent to Claude to enrich
The start of a note, right after you save it
When you create or edit a note, Cyton sends only its opening - about the first 600 characters (~100 words) - to Claude, to file it under a memory tier, tag it, and suggest links. The rest of the note stays on your Mac.
More of a note when it's summarised
Up to a few thousand characters when a note is condensed for review or scanned for embedded concepts.
Trimmed session messages
Short excerpts of your Claude sessions, so Cyton can spot recurring themes.
Titles only, for some jobs
Connection-finding and the review filter send note titles and signals - never the body.
You see every node as it goes
This happens the moment you save a note - never on a hidden schedule. Only the opening of that one note goes out (the first ~600 characters), never the whole note. The app's activity feed names each node as it's enriched ("AI enriched: my-note → semantic"), so nothing is sent silently.
Your graph grows itself - and AI writes the new nodes
Cyton doesn't only tag what you wrote. As it works, it spots concepts worth their own node, creates them, and writes their content with AI (Claude) - so a good part of your graph is machine-generated, not typed by you. Those notes are clearly marked as Cyton-generated ("Built by Cyton"). The notes you wrote yourself are never rewritten.
Free plan
Those note openings go to Claude through Cyton's own Anthropic account - we pay for it, capped at 500 enrichments a day.
Pro plan
Enrichment routes through your own Claude via Claude Code. Cyton's servers and API key are never in the loop.

No note content is ever stored on a Cyton server, sold, or used to train a model. Crash reports are stripped of note content before they are sent.

Obsidian sync

Your vault.
Cyton watches it.

Point Cyton at your Obsidian vault and walk away. Every save is processed instantly - concepts extracted, nodes created, synapses updated. No plugins, no manual steps, no disruption to your existing workflow.

Works with any vault size
Watches for changes in real-time
Non-destructive - never edits your files
Activity feed
node created: "distributed cognition" → sensory
synapse strengthened: startup-strategy ↔ cognitive-load (0.71)
semantic tier: "knowledge management" consolidated from 8 episodes
review queued: "spaced repetition" - 14 days stale
REM suggestion: PKM-workflow ↔ habit-formation (94%)
microglia: deferred "old-todo" - retrievability 0.02
Under the hood

Curious how it actually works?

Go under the hood - the forgetting curve, the signals Cyton scores every note on, and how the graph rewires itself as you think.

Find out how it works
Pricing

Simple, honest pricing.

Free forever for individuals. Pro unlocks the full neural stack.

Free
$0
Forever

Download the app to get started. No account needed until you're ready to upgrade.

500 neurons
Full knowledge graph
Obsidian vault sync
5 memory tiers
Spaced repetition queue
Auto enrichment - Haiku API
Cyton's cloud enriches your notes automatically
1 Claude Code session / day
Pro
Most popular
$12/mo
Billed monthly

Connect your own LLM via MCP. All enrichment runs locally - your data never leaves your machine.

Everything in Free
Unlimited neurons
MCP enrichment - your own LLM
Sonnet, GPT-4, Gemini - runs entirely on your machine
Unlimited Claude Code sessions
REM pattern completion
Nightly graph consolidation
Multi-vault support
Graph data export
Priority support
Your data stays yours
Search and indexing run on-device - your notes aren't uploaded to be searched, and the MCP bridge is loopback-only. Enrichment sends only the opening of each note to Claude (Cyton's account on Free, your own on Pro) - nothing is stored on our servers, sold, or used for training. Crash reports are scrubbed of note content.
Local first
Crash reports only
Open graph format