whitepaper

The Agent-Native
Personal OS

How fyltr rethinks the relationship between humans, agents, and data

fyltr.ai team on behalf of humanity
March 2026

the mission

Let humans be more human

by freeing them from the prison of deception,
misinformation, and digital noise

The most human thing is love. And that's the only thing we need in the age of AI.

— LinkedIn, Mar 2026

the noise — processed reality

“Sugar makes the body feel good — without nourishing it. Social media makes connection easy, but often leaves us exhausted — and the relationships rarely last.”

“Two optimized personas exchanging perfectly crafted messages is not really a relationship — it's algorithmic role-play.”

“We risk building a world that looks incredibly connected on the surface… but becomes increasingly empty underneath.”

the signal — distilled knowledge

“The real skill today isn't access to information. It's the ability to tell the difference between what feels good now and what is actually good for you.”

“AI should help us cut through noise. Help people find genuinely compatible partners. Help us connect with products and ideas that are intrinsically meaningful.”

“Maybe happiness isn't about rejecting technology. Maybe it's simply about being free to shift attention between codes and frequencies. Between mind and heart.”

the agency

human agency — the free spirit

A human has intrinsic will. The capacity to choose, to refuse, to create meaning from nothing.

Free will is not a feature. It's the foundation. Everything else — tools, systems, organizations — is built on top of it.

But when thousands of people act within the same incentive system, something strange happens. The system starts shaping the decisions of its participants.

system agency — the corporation

A corporation behaves like an organism. It has goals. It makes decisions. It remembers. It adapts. No single person inside fully controls it — yet the system clearly acts.

Economists call this emergent agency. The organization becomes a decision-making machine built from incentives, rules, culture, and information flows.

But unlike humans, corporations don't have intrinsic will. They only have structured human will. A corporation is a machine that aggregates and amplifies human intentions.

And now AI enters the picture. When AI agents start operating inside organizations — making decisions, optimizing processes, allocating resources — something new may emerge:

Not just collective intelligence. But synthetic institutional agency.

The question is no longer whether machines will have agency.
It's whether systems composed of humans and machines will develop a will of their own.

“AI extends human cognition just as tools extended our hands and machines extended our muscles. But AI itself is not an independent entity. It's a layer in the stack of intelligence, not the origin of it. And we're still the ones writing the first lines of code.”

— LinkedIn, Mar 2026

The cost of a thing is the amount of life you exchange for it.

Henry David Thoreau

Attention is the rarest and purest form of generosity.

Simone Weil

02 — the premise

How AI agents actually work

💬

User Intent

Natural language instruction or goal

🧠

LLM Reasoning

Foundation model plans & decides

🔧

Tool Execution

APIs, databases, code, actions

Outcome

Task completed, result returned

An agent is an LLM in a loop: it receives a goal, reasons about what to do, executes actions through tools, observes results, and iterates until done. The loop is simple. What makes an agent useful is something else entirely.

03 — the bottleneck

An agent is only as good as
its three pillars

1

relevant data

Context

Without the right data at the right time, even the smartest model hallucinates. Most agents fail here — they lack access to your emails, messages, contacts, documents, health records, and the relationships between them.

Context is the #1 bottleneck in real-world agent performance.

2

intelligence

Capabilities

The model's reasoning ability determines what it can do with context. Claude, GPT-4, Llama — each brings different strengths. But raw IQ without data access is like a genius locked in an empty room.

Capabilities are necessary but not sufficient.

3

experience

Training

System prompts, skills, and learned patterns shape how an agent behaves. A well-trained agent knows your preferences, your communication style, which contacts matter, and how you organize your life.

Training is what makes an agent yours.

4

underlying free will

The Prompt

The prompt is the soul of the agent — its underlying intent, values, and direction. Without a well-crafted prompt, an agent has no purpose, no judgment, no free will.

The Prompt is what gives an agent autonomy.

Agent Performance = Context × Capabilities × Training × Prompt

04 — the architecture

How fyltr solves this

powered by angee.ai
Every domain exposed twice: REST APIs for humans, MCP tools for agents. Same data, same logic, two interfaces.
user → rest api&agent → mcp tools
UI / UX Layer
📥Unibox
👥Nexus
🔗Connect
🤖Agents
📚Knowledge
🏥Health
🔔Attention
📋Projects
🎓Learn
rest · websocket · openapi
MCP Tools Layer
list_messagesget_messagelist_threadssearch_contactslist_contactsget_contactget_health_summarylist_identities
tool calls · structured I/O
Agents — ANGEE Runtime
Phill
Claude Code
email triagescheduling
Researcher
Letta
web searchsynthesis
Health Coach
OpenCode
vitalsFHIR
Custom
Any LLM
your skillsyour tools
django orm · celery tasks · websocket channels
Apps — Business Logic
📥Unibox
MessageThreadIdentityPlatformPartFragment
👥Nexus
ContactTopicGroupRelationshipEmbedding
📚Knowledge
DocumentChunkEmbedding
🏥Health
ObservationConditionInterventionGoal
🔗Connect
AccountSecretMCPServerWebhook
🤖Agents
AgentTemplateSkillToolProvider
🔔Attentionsoon
FeedSignalFilterPriority
📋Projectssoon
ProjectTaskMilestoneBoard
🎓Learnsoon
CourseLessonProgressQuiz
sql · pgvector · fernet encryption
Data Layer
🗄️

Relational + Vector

PostgreSQL 17 + pgvector

All domain models & relationships

Full-text search (tsvector)

Semantic search (embeddings)

HNSW index for ANN queries

Advisory locks for sync

Audit trail (simple-history)

🕸️

Social & Knowledge Graph

Django ORM graph models

Contact relationships (follows, knows, works_with)

Topic hierarchy (leaf / branch / root)

ConceptLinks between topics

Quotation-based threading graph

Embedding-first topic resolution

Contact dedup & merge tracking

📁

File & Object Storage

Google Cloud Storage + local

Document uploads (PDF, notes, pages)

Message attachments (MIME parts)

Health records (lab reports, scans)

Agent skills (.md in git repos)

Fernet-encrypted credentials

Self-hosted option (data on device)

05 — the result

Your data. Your agents.
Your operating system.

🔒

Self-Hosted

Your data never leaves your infrastructure. Run on your own hardware or cloud.

🔌

Open Protocol

MCP means any model can plug in. Not locked to one vendor. Swap Claude for Llama overnight.

🧩

Composable

Mix agents, skills, tools, and data sources. Build personal workflows that compound over time.

🌐

Open Source

Fully transparent. Audit every line. Extend anything. The community builds together.

Alexis Yushin · fyltr.ai · March 2026