Your life has an
operating system now.
fyltr is your personal agentic OS. It installs the ontology of your life — contacts, conversations, documents, health records, your calendar — as typed, permission-scoped objects that AI agents can actually work. Not a chatbot with your inbox pasted in.
The OS gives agents hands. The ontology gives them a world. fyltr installs the world of you.
self-host on ANGEE • private beta • your data, your models — down to fully local
A world, not a chatbot
Agents operate typed, permission-scoped objects — your contacts, messages, documents, health records — not text you paste into a prompt. A world they can reason over and act on.
You choose what AI sees
Every file, model, and agent carries a clearance. A provider only ever sees what its ceiling allows — down to running a local model for your most sensitive data and nothing in the cloud.
Your life, imported
Bring in real EHR exports, your calendar, your contacts via CardDAV, email and WhatsApp. fyltr types it all into one live world instead of scattering it across a dozen apps.
Built on an open foundation
fyltr runs on ANGEE, the open-source (AGPL-3.0) agentic OS. The foundation is open and self-hostable, so the OS answers to you — not to a vendor.
How a personal agentic OS works
An agentic OS gives agents what a traditional OS gives programs: identity, permissions, tools and audit. fyltr adds the one thing that makes them useful — a world of your own.
Ontology
Your life, typed
fyltr installs a domain ontology for a life. Contacts, conversations, documents, health records, calendar events, projects — each becomes a typed object with real fields and relationships, not a wall of text.
- ✓Contacts, relationships, groups and topics (with CardDAV sync)
- ✓Documents with pipelines, embeddings and a folder graph
- ✓Health: imported EHR data — labs, medications, vitals, visits
- ✓A unified inbox for email and WhatsApp
The OS is domain-blind. The ontology gives agents a world.
Agents
Hands on your world
An MCP server exposes your world to agents as permission-scoped tools across files, contacts, health, messaging and search. Agents don’t guess at pasted text — they call typed operations that answer to you.
- ✓MCP-native tools, each scoped to a permission and a clearance
- ✓Works with Claude Code and OpenCode over ACP
- ✓Stateful agent memory so agents remember across sessions
- ✓Bring your own inference: Anthropic, OpenAI, Mistral, or local Ollama
Agents operating a typed world are employees. Agents guessing at text are liabilities.
Clearance
Sovereignty, enforced
Every file, user, provider, model and agent carries a clearance — EXTERNAL, CONFIDENTIAL or SECRET. A provider’s clearance caps what it can ever see, cascading provider → model → template → agent and enforced at serving time.
- ✓Run a local model for SECRET data; let cloud providers see only EXTERNAL
- ✓Metadata scrubbed — filenames, emails and URLs never enter prompts
- ✓A manual redaction pipeline produces a reviewable redacted draft
- ✓Embeddings run locally by default
You choose which AI sees what — down to running it all locally.
What agents can do with a world
Real verticals, shipping today. Health leads — it is the most mature part of the OS.
Your health records, finally legible
Import real EHR exports — Epic/MyChart, Elation timelines and clinical profiles. Labs, medications and vitals become typed objects an agent can read, chart and cross-reference. Ask questions about your own history instead of digging through PDFs.
Documents that answer back
Files are chunked, embedded and organized in a folder graph, then classified by clearance. Ask across everything you own — and when something has to leave, the redaction engine blacks out your word lists on PDFs and images for a reviewable draft first.
One inbox for the people who matter
Email over IMAP and WhatsApp land in one unified inbox, tied to the contacts and topics in your world. More channels are on the roadmap — but what ships today is real, not a mockup.
A relationship graph, not a rolodex
Nexus keeps your contacts, relationships, groups and topics as first-class objects, synced over CardDAV. Agents reason about who someone is and how you know them, then act with that context.
Your coding agent, on your life
Point Claude Code or OpenCode at fyltr over ACP and it works your world through ~76 permission-scoped MCP tools — files, contacts, health, messaging, search — each capped by clearance. Stateful memory means it remembers between sessions.
SECRET stays home
Give your health folder a SECRET clearance and route it to a local model — no cloud provider can ever see it. Let EXTERNAL email drafts use a frontier model. One rule, enforced everywhere at serving time.
Sovereignty is a system, not a slogan
The clearance system is fyltr’s signature feature — and it is real, tested, and enforced everywhere. You choose which AI sees what, down to running it all locally.
Clearance on everything
Every file, user, provider, model and agent carries a clearance: EXTERNAL, CONFIDENTIAL or SECRET.
Provider ceilings that cascade
A provider’s clearance caps what it can ever see — cascading provider → model → template → agent, enforced at serving time.
Local models for secret data
Run a fully local model for SECRET data and let cloud providers see only EXTERNAL. Nothing sensitive leaves by accident.
Metadata scrubbing
Filenames, email addresses and portal URLs never enter agent prompts. Tested, not aspirational.
Manual redaction pipeline
Define word lists; OCR blackout removes them from PDFs and images, producing a reviewable redacted draft before anything ships.
Local-by-default embeddings
Vectors are computed on your own hardware by default (Ollama, nomic-embed-text) — your corpus stays yours.
fyltr is designed for privacy from the architecture up. It is early software and makes no compliance or certification claims — the clearance system is how sovereignty is enforced, not a promise on a page.
Read the trust & architecture story on ANGEE →Built on an open foundation
fyltr is the personal edition of ANGEE, the sovereign agentic OS. The OS gives agents hands. The ontology gives them a world.
An open foundation
ANGEE is the open-source (AGPL-3.0) agentic OS: identity, permissions, addons, an MCP system-call surface and an audit log. It is building in the open — early alpha, same team.
- •AGPL-3.0 licensed
- •Self-hostable
- •MCP-native, addon-based
fyltr is an edition
The OS is domain-blind. Editions install a domain ontology on it. fyltr installs the ontology of a life; the foundation is open, but the fyltr edition is a product.
- •Personal ontology
- •Not itself open source
- •Where ANGEE began
ARPEE is its sibling
ARPEE is the enterprise agentic OS — the open-source agentic ERP for companies. Same framework, a different world: fyltr installs you, ARPEE installs the company.
- •Enterprise edition
- •Open-source agentic ERP
- •Built on ANGEE
What plugs into your world
fyltr imports the real sources of your life and types them into one ontology. This is what ships today — more channels are on the roadmap, and we will not pretend otherwise.
Health imports
The flagship — and a real differentiator- ✓Epic / MyChart (C-CDA / XDM)
- ✓Elation timeline + clinical profile
- ✓Labs · Medications · Vitals
Messaging
Live today- ✓Email (IMAP sync)
Calendar
New- ✓Google Calendar
- ✓Outlook
- ✓iCal
Contacts
Sync- ✓CardDAV
Documents
Pipelines + embeddings- ✓PDF · Office · Images (OCR)
- ✓Folder graph + classification
Inference
Bring your own- ✓Anthropic · OpenAI · Mistral
- ✓Ollama (local) · Letta · custom
Calendar and projects are new and early — we describe them as such. More messaging channels are on the roadmap, not the feature list.
Own it, or let us run it
Self-host free on the open foundation, or join the waitlist for managed fyltr.ai.
Self-host
On ANGEERun the fyltr edition on ANGEE, the open-source agentic OS
- ✓Deploy on your own hardware (Docker Compose)
- ✓Postgres + pgvector, Redis included
- ✓Bring your own inference — including fully local
- ✓Local-by-default embeddings
- ✓The full clearance system
Managed
fyltr.aiWe host and run your personal agentic OS
- ✓Managed deployment at fyltr.ai
- ✓Guided EHR, contacts and document import
- ✓Clearance-based provider ceilings, set up for you
- ✓Unified inbox: email and WhatsApp
- ✓Early access via the waitlist
Concierge
White-gloveHands-on onboarding for your most sensitive world
- ✓White-glove setup and migration
- ✓Local models for SECRET data
- ✓Custom ontology and import work
- ✓Dedicated support
- ✓Household / family scope
Frequently asked questions
Everything you need to know
What is a personal agentic OS?
What data does fyltr hold?
Which AI providers can I use? Can it run fully local?
How does fyltr keep my data private?
Is my health data safe?
How does fyltr relate to ANGEE and ARPEE?
Is fyltr open source?
Can I self-host?
Install the world of you
fyltr is in private beta. Join the waitlist and be among the first to give AI agents a typed world of your own — on your terms.
Prefer to own the whole stack? fyltr is built on ANGEE, the open-source agentic OS.
Explore ANGEE↗