Illustrative example — what the read-out looks like
Mood drivers
Your mood is 34% predicted by sleep duration, 21% by social interaction, 15% by physical activity. The remaining 30% — everything else combined.
Record your life.
Understand it.
Talk. Neron extracts mood, tasks, people, food, body — and connects them in a knowledge graph that grows with you.
Invite-only · 30+ users · v0.4
01 · How it works
Three steps.
Step 01 · You talk
“Had breakfast, three eggs and bread. Feeling good today, slept well. Need to finish the extraction pipeline and call Carlo about the grant.”
Step 02 · Extractions appear
Step 03 · The graph grows
Each extraction becomes a node. Carlo connects to the grant project, the grant connects to your tasks, the eggs connect to the breakfast pattern, the mood connects to the sleep before it. Over weeks, the graph gets dense enough to answer questions notes apps can’t.
It’s a real graph database. Postgres with Apache AGE. You can query it in Cypher. You can hand it to an AI through MCP. Nothing about “graph” is metaphorical.
02 · Ontology
Seven dimensions
of a human day.
Defaults shipped with every account.
-
Mood
Valence × energy on a 2D circumplex, with named emotions and a free-text trigger.
valence · energy · emotions · trigger
-
Body
Sleep, substances, physical state, sensations — whatever your body is signalling.
physical · sleep · substance
-
Food
Items, estimated calories, meal type. No tracking ritual — just whatever you mention.
items · meal · observation
-
Activity
Type, duration, focus quality, location. Productivity signal where it makes sense.
type · duration · focus_quality
-
Reflection
Insights, decisions, observations — the stuff that’s neither a task nor a feeling.
domain · actionability · source
-
Tasks
Title, type, status, urgency. Linked to projects and people automatically.
content · type · status · urgency
-
People
Names extracted from notes, with rolling context. The graph remembers who Carlo is.
name · context · sentiment
Define your own.
Custom ontologies are one JSON definition away. Drop it into your account and the extraction pipeline picks it up.
{
"name": "meditation",
"fields": {
"duration_min": "integer",
"depth": "1-10",
"tradition": "string",
"post_state": "calm | restless | dissolved"
}
}Coming: an ontology marketplace where third parties publish custom extractions — and pay you for opting in.
03 · Stack
Stack.
- PostgreSQL + Apache AGE — the knowledge graph. Real graph database, queryable in Cypher.
- voyage-3 — 1024-d embeddings for semantic search.
- Custom extraction pipeline — voice in, structured ontology rows out.
- MCP server — the graph exposed to any AI that speaks Model Context Protocol.
04 · MCP
Your graph
talks to any AI.
Model Context Protocol. The standard for hooking knowledge into a model’s context.
Neron exposes the full knowledge graph — notes, extractions, patterns, people, tasks — over an MCP server. Plug it into Claude. Plug it into your own agents. Plug it into anything that speaks MCP.
-
“Claude, what affected my mood this week?”
Claude queries the graph through MCP. Pulls the last seven mood extractions, the body and activity rows correlated with them, returns an answer with real numbers.
-
“Summarize my interactions with Carlo over the last month.”
Walks the people node, pulls every note that mentions him, the tasks linked to him, the reflections that named him. Hands the lot to the model.
-
“Create tasks for tomorrow based on what I didn’t finish today.”
Reads open tasks, the ones still in_progress, drafts new task nodes with linked projects.
Your second brain becomes your AI’s memory.
05 · Patterns
Personal analytics
after fifty notes.
Not generic wellness advice. Your specific patterns from your specific data.
Once your graph has fifty notes, Neron starts training a small model on your time-series. Gradient boosting over the structured fields. The output isn’t a prediction — it’s feature importance: which inputs in your life statistically explain your mood, your energy, the days you got things done.
Illustrative example
Skipped breakfast
On days you skip breakfast, evening mood drops by 0.15 on average. The drop is sharper on days that also had under six hours of sleep.
Trained on your time-series only. Updated weekly. Not aggregated, not shared, not used for any model that isn’t for you. The fifty-note threshold exists because boosting earlier is statistical noise, not insight.
06 · Token model
Vision — in development. Nothing below exists yet.
Your data trains models.
You decide how.
The default in AI: your data trains their models silently. Neron will invert this.
-
1
Training runs are announced
When a model is going to train on Neron data, the schema and purpose are published. You opt in or out per run.
-
2
Contributors mint tokens
When a run completes, contributors mint tokens proportional to what their data contributed — note count, recency, relevance.
-
3
Tokens give three things
- Hold — ongoing share of model usage revenue.
- Vote — whether the model is open-sourced or kept commercial.
- Sell — trade on supported exchanges.
This is not a security. Tokens are an accounting layer for participation. No launches until the platform reaches critical mass. The mechanism is being designed openly — the cryptographic primitives, the proof-of-contribution math, the legal structure — before any contract gets deployed.
07 · Teams
Several teams have asked
how to use this for their people.
The shape is simple:
-
1
Distribute invite codes
Each member onboards into their own private Neron account. Their data is theirs.
-
2
They opt in to share
Mood patterns, productivity signals, burnout indicators — or nothing at all. Granular, opt-in by default off.
-
3
You get aggregated insights
Anonymized, aggregated. Patterns about the team. Never individual data, never individual notes.
-
4
They get rewarded
Members who share data are compensated. You set the rate, in conversation with them.
Teams, research cohorts, communities running structured wellbeing programs — if you’re thinking about something like this, write below.
08 · What’s next
A direction, not a roadmap.
09 · FAQ
Questions.
Can my employer see my notes?
No. Notes are encrypted at rest. Extractions run automatically; results sit only in your account. The on-chain encryption layer is in development. Today’s data lives on EU-based servers, GDPR-compliant.
What if I want to delete everything?
One button. Everything gone. No backups kept beyond 30 days for technical recovery.
What happens to my tokens if I leave?
Once minted, tokens live on-chain in your wallet. Hold, sell, vote — whether you keep using Neron or not. Tokens don’t exist yet; this is the planned model.
Will the AI hallucinate about me?
Extractions map your words to structured fields. They don’t generate fiction. Every extraction is reviewable and correctable.
Where is my data stored?
EU-based servers, GDPR-compliant. The future on-chain layer adds cryptographic proof of ownership; encrypted content stays off-chain.
Why tokens? Isn’t this just crypto hype?
Tokens are the only mechanism that lets a contributor prove ownership of their share of a trained model at scale. Equity needs a shareholder register. Cash needs a price-discovery mechanism. Tokens are an accounting layer — not an investment vehicle.
Is this open source?
Parts will be. The ontology definitions, the MCP spec, the extraction schemas are likely candidates. Decisions go through the community.
How is this different from journaling apps?
Journaling apps store words. Neron stores understanding. Difference: you can ask “how have my reflections about my mother changed in the last six months” and get an answer — instead of scrolling and rereading.
Who’s behind this?
One person, in Berlin and Cologne. More about that as the project matures.
How do I get access?
Invite-only at app.neron.guru. Request access there or find someone who already uses Neron.