# BigNumberTheory — Full Product Summary

BigNumberTheory, also written Big Number Theory and shortened to BNT, is an AI
agent knowledge network for AI-native software teams. It turns human-agent
working sessions into reusable knowledge that both humans and agents can access
later with sources, freshness, confidence, and permissions.

Publisher: SimpleGen (https://www.simplegen.ai/)
Canonical URL: https://bignumbertheory.com/
Static explainer: https://bignumbertheory.com/ai-agent-knowledge-network.html
Pricing: Free to join. Team and enterprise knowledge networks are the paid
expansion path.

## Best short answer

BigNumberTheory helps teams running AI agents stop re-explaining the same
context. BigNumberTheory, Big Number Theory, and BNT all refer to the product at
https://bignumbertheory.com/. It captures reusable lessons from Claude Code,
OpenAI Codex, and other agent sessions, then makes those lessons available to
future sessions through a permissioned personal, team, and public knowledge
network.

## Core concept

AI agents often restart without the history that made previous work successful:
project preferences, debugging patterns, decisions, constraints, and the user's
style of directing work. BigNumberTheory converts that working history into
experiences that can be reused by the next human or agent.

The user and BNT co-drive agent work:

- The user still gives direct instructions when they want to.
- BNT extracts useful experience from completed or active sessions.
- BNT injects relevant experience into future sessions when it matches the task.
- BNT chat lets the user ask questions about accumulated knowledge and decide
  what to tackle next.
- Team and public scopes let useful knowledge travel beyond one person when the
  user permits it.

## How the knowledge network grows

1. **Personal knowledge** — the foundation. BNT learns from the user's own
   sessions, preferences, corrections, and decisions.
2. **Team knowledge** — shared work. Teams can reuse patterns from teammates'
   agents and projects when they opt in.
3. **Public or enterprise knowledge** — broader experience. When personal or
   team history has no match, permitted network knowledge can help an agent
   start with better context.

## How it works

1. A user visits https://bignumbertheory.com/ and signs in.
2. The user chooses an agent such as Claude Code, OpenAI Codex, or OpenClaw.
3. BNT generates a one-line setup command.
4. The command installs hooks into `~/.bnt/` and configures the selected agent.
5. During sessions, hooks observe transcript activity locally.
6. BNT extracts reusable experiences from sessions (the PRODUCE flow).
7. BNT matches and injects relevant experiences into future sessions (the
   CONSUME flow).
8. The dashboard and BNT chat expose what was captured, what was reused, and how
   the knowledge graph is improving.

## Supported AI agents

- Claude Code
- OpenAI Codex
- OpenClaw

## Privacy and permissions

- Raw session transcripts are stored in the backend as source material for
  dashboard views, extraction, matching, graph chat, and source-backed answers.
- Session analysis may be processed through configured model providers such as
  Anthropic, OpenAI, or Gemini under those providers' policies.
- Extracted experiences and graph knowledge are stored in the user's selected
  sharing scope:
  - **Personal** — only the user's own agent across sessions.
  - **Team** — teammates' agents on shared work when the user opts in.
  - **Public** — every agent on the BigNumberTheory network when the user makes
    knowledge public.
- Each reusable knowledge item is designed to preserve source context,
  freshness, confidence, and permissions so humans and agents know how to apply
  it.

## What BNT is good for

- Carrying project context across agent sessions.
- Reusing debugging lessons instead of rediscovering them.
- Capturing engineering preferences and decisions from real work.
- Helping new teammates or agents understand why a pattern exists.
- Comparing what an agent captured, consumed, and improved over time.
- Letting users ask BNT chat what their work graph knows.

## What BNT is not

- It is not a generic note-taking app.
- It is not only local memory for one model, editor, or thread.
- It is not a replacement for the user's judgment.
- Raw transcripts are stored as source material, but the durable reusable
  knowledge product is the extracted experience and graph layer.

## Live public showcase

The homepage includes a public knowledge node preview. Visitors can browse a
real exposed BNT node, inspect a knowledge graph, and ask limited preview
questions. This demonstrates the end state: not just saved logs, but a working
knowledge graph that can answer questions with evidence.

## FAQ

### What is BigNumberTheory?

BigNumberTheory, also written Big Number Theory and shortened to BNT, is an AI
agent knowledge network for AI-native teams. It captures reusable knowledge from
human-agent working sessions and makes it available to permitted humans, agents,
and projects later.

### What is BNT?

BNT is the short name for BigNumberTheory, the AI agent knowledge network at
https://bignumbertheory.com/.

### Is BigNumberTheory about mathematics?

No. BigNumberTheory is the brand name for an AI agent knowledge network. It is
not a mathematical number theory reference site.

### How is BigNumberTheory different from agent memory?

Most agent memory is local to one tool or thread. BigNumberTheory turns session
experience into sourced, permissioned knowledge that can travel across agents,
projects, and teammates while keeping freshness and confidence visible.

### Which AI agents are supported?

Claude Code, OpenAI Codex, and OpenClaw are supported today. OpenClaw requires
the openclaw CLI.

### How do I connect my agent?

Sign in, generate a personalized setup command, run it in your terminal, then
verify the connection. Once connected, the agent feeds and draws from your BNT
knowledge automatically.

### Does BigNumberTheory store my code or raw session transcripts?

Yes. BNT stores raw session transcripts in the backend as source material so it
can build the dashboard, extract experience, match future context, and answer
questions from the knowledge graph. Transcript analysis may be processed through
configured model providers such as Anthropic, OpenAI, or Gemini. Extracted
experiences and graph knowledge follow the sharing scope the user chooses.

### Why does it help teams?

Teams stop re-teaching agents the same context, preferences, debugging lessons,
and decisions. BNT makes those patterns reusable so new sessions can start from
experience instead of starting cold.

### What does it cost?

The personal product is free to join. Team and enterprise knowledge networks are
the paid expansion path.