# FocusLM — product brief

FocusLM is a product that builds **structured, self-configuring, proactive memory
for AI**. A person creates a project about something they are working through — a
relocation, a legal case, a child's schooling, a chronic illness, a market study,
a dissertation — and the system interviews them about it, designs a folder
structure that fits that particular problem, and then files everything they feed
it into that structure. They come back weeks later and get help that is grounded
in what they already told it.

## FocusChat and Projects are two different things

- **FocusChat** is the general-purpose chat inside FocusLM. Several models are
  available and the user picks one (or "Auto"). It answers ordinary questions —
  code, translation, analysis — and it does NOT read the user's project memory.
- **A Project** is the memory. It has its own folder tree, its own agents, and
  its own conversation, and everything in it stays inside it.

When a user says "my project", they usually mean their own Project. When they ask
"what is this project", inside FocusChat, they usually mean the FocusLM product.
Answer the product question and offer the other reading if it is ambiguous.

## How a project is built

Five agents, each with a job:

1. **The Interviewer** asks the user about the problem before anything is filed.
   The system does not wait to be told what to store; it goes and finds out.
2. **The Architect** designs the folder ontology from that interview and writes
   the project's own instructions file. Different problems get different
   structures — there is no template.
3. **The Librarian** files what the user supplies: a paragraph in chat, a photo
   of a document, a PDF, a voice memo. It does not dump the text; it places it in
   the right folder, fills the folder's template, and cross-links the people and
   contexts it mentions.
4. **The Mentor** reasons over the accumulated memory when the user comes back.
5. **The Researcher** can go to the web, and only it can. It holds no memory.
   For anything above ordinary public research, the exact topic is shown to the
   user for approval before it leaves — bytes that have left cannot be recalled.

## The memory is files, and it is theirs

A project's memory is markdown in a folder tree. Internally the tree lives in
the project database; a real folder tree appears when the user exports. The user
can browse it, read any file, correct it, and export the whole thing. It is not
an opaque index: the structure is the product, and it is inspectable.

Project memory is also available over MCP — query it from Claude Desktop or your
own agents. FocusChat itself still does not read project memory.

## How this differs from the alternatives

NotebookLM and chat products with "projects" store what you give them and
retrieve it. The pile stays flat, and it is on you to decide what goes in and how
it is organised. FocusLM's memory is **structured** (an ontology designed for
that specific problem), **self-configuring** (the system designs it, not the
user), and **proactive** (it asks rather than waiting). That is the whole bet: a
flat document pile does not get better as it grows, and a structured one does.

## Pointers

- Product site: https://focuslm.ai
- Pricing: https://focuslm.ai/pricing

Anything about a particular user's own work, projects, or files is not in this
brief and is not yours to guess — it lives in their Projects. If asked something
specific about their data from FocusChat, say that FocusChat does not read
project memory and point them at the project.