FocusLM builds a living, structured memory of raising your child.
FocusLM interviews you, shapes a deep, living memory around your problem — its own organization, its own instructions — and files everything you feed it, so every answer is grounded in your accumulated reality, not a generic reply. Nothing to set up, no prompts to write. Ever.
Runs on the frontier models you trust
Memory, not a document pile
Four things make FocusLM different from a folder of files or a long chat thread.
Structured
Your material is organized and cross-linked — not a flat pile. That's what makes grounded citation and cross-context reasoning possible at all.
Self-configuring
The agent designs the memory and writes its own operating instructions. You get a developer-grade setup without being a developer.
Proactive
It interviews you up front and keeps the memory evolving as you feed it — instead of waiting for you to organize everything yourself.
Searchable by meaning
Everything you file is semantically indexed, so the right passage surfaces by meaning — precise, cited retrieval over an organized memory, not a keyword grep.
A large structured memory beats a long chat context
Stuffing everything into a chat window stops scaling. Grounded, structured memory keeps pulling ahead the more you feed it.
The research behind these numbers: Lost in the middle·Memory like an OS·RAG vs long context·More context isn't better
How FocusLM works
You never organize anything or write a prompt. That is the system's job.
- 1
It interviews you
The agent opens the conversation. It asks what you are investigating, who is involved, and what a good outcome looks like — until it can commit to a shape for your memory.
- 2
It designs the memory
FocusLM shapes the memory itself — how it's organized, the frameworks the agent reasons with, and its own operating instructions. You approve it once.
- 3
You feed it, it files
Drop in a chat message, a photo of a note, a PDF, a voice memo, or a whole folder as a .zip. Each one is filed exactly where it belongs and cross-linked to the people and contexts it touches.
- 4
It reasons, grounded
Come back any time and it answers from your own memory — citing it — and spots patterns across contexts. As your life changes, the memory grows and deepens with it.
- 5
It enriches — when you allow it
When a question needs outside facts, an agent searches the web behind an allowlist and your approval — you see the exact topic that leaves — and files the cited result back into your memory.
One system, five agents
Each agent does one job — together they turn what you feed in into memory.
The Interviewer
You describe the problem in your own words. The Interviewer proactively asks what you're investigating, who's involved, and what a good outcome looks like — until it can commit to a shape for your memory.
- Opens the conversation
- Asks until it understands
- In your language
The Architect
From that brief it designs the memory itself — how your material is organized and the memory's own operating instructions — and asks you to approve it once.
- Designs the folder ontology
- Writes the memory's own guide
- You approve it once
The Librarian
Feed it anything over time — a chat, a PDF, a photo of a note, a voice memo, a whole folder as a .zip. Each item is extracted and filed exactly where it belongs, cross-linked and versioned.
- Chat, PDF, image, voice, .zip
- Filed to the right place
- Cross-linked and versioned
The Mentor
Come back any time and it answers from your own memory — citing the notes it used — and spots patterns across contexts you'd never connect by hand.
- Cited, grounded answers
- Cross-context patterns
- @-mention any note
The Researcher
When a question needs outside facts, it searches the web behind a per-project allowlist and your approval — you see the exact topic that leaves — and files the cited result back into memory.
- Searches, never silently fetches
- Cited documents
- You see what leaves
FocusChat — a fast chat across every model
Not every question deserves a whole project. FocusChat is a quick, general-purpose chat over the leading models — switch models mid-thread, attach a file, dictate with your voice, generate an image. It sits beside your project memory, so a throwaway question never clutters it.
- Switch between the top models
- Voice, image and file attachments
- Kept separate from project memory
Built for how real material arrives
The practical parts that make the memory usable day to day.
Attach to Claude Desktop
Your project memory is available over MCP — query it from Claude Desktop or your own agents.
Voice, images, any format
Dictate a note, drop a photo, a PDF, a spreadsheet, or Markdown — it's extracted and filed.
Whole folders as .zip
Upload an archive and it's unpacked, deduplicated, and filed piece by piece.
@-mention your memory
Point the agent at any note or folder mid-conversation for a grounded answer.
Roles & sharing
Invite people to a project with the access you choose; every change is attributed.
Export & versioning
Plain markdown you can download; every edit is versioned and inspectable.
Built for the problems you carry for years
No two projects are alike — each gets its own memory, shaped around it.
Raising a child
Track routines, incidents, and what actually works across home, school, and activities.
A chronic illness
A dated log of symptoms, treatments, and questions — reasoned over, not lost in a notebook.
A legal case
Parties, threads, and dated events, filed and cross-linked so nothing slips.
A relocation
Options, constraints, and decisions in one connected place as the plan evolves.
A market or PhD
Frameworks, sources, and insights that compound into real understanding.
A hiring process
Candidates, signals, and decisions — structured, dated, and grounded.
Private by design
Your most sensitive material — a child, an illness, a legal case — stays yours.
Built on Anthropic + OpenAI
Your content is processed by frontier models under terms that forbid training on it.
Isolated per project
Each project's memory is fully separate — your material never crosses into another.
EU AI Act aligned
Transparent, inspectable memory and compliance presets, not a black box.
Yours to export
Plain markdown you can read and download any time. No lock-in.
Questions
The short version of how FocusLM is different.
Do I have to organize anything or write a prompt?
No — that is the entire point. The system interviews you and builds the memory and the agent's instructions for you. You just approve the result.
How is this different from NotebookLM or Claude Projects?
They store documents. FocusLM turns them into a living memory the agent configures and maintains, which is what makes grounded, cross-context reasoning possible.
What can I feed it?
Chat, notes, PDFs, images, spreadsheets, Markdown — even a .zip of a whole folder. Each is extracted and filed into the right place automatically.
Where does my data live?
Your memory is a set of plain, private files — yours to inspect, and every change is versioned. Projects are fully isolated — your material never crosses into another project.
Do I need to be technical?
No. FocusLM gives non-developers the kind of structured, self-configuring setup that used to require a terminal and knowing how to design a system for yourself.
Bring the problem. FocusLM builds the memory — and the method to work it.
Start your first project — no setup, no prompts to write. Ever.