# Neither — architecture overview

**Version 1.2 — the context layer as a product**

## The problem

Organizations generate decisions, commitments, and relationships across dozens of channels. That intelligence is scattered and unstructured. AI agents proliferate, but they lack **persistent organizational context** — every session starts from scratch.

The missing infrastructure is a **context layer**: structured, persistent intelligence that both humans and agents can query.

## Four layers (conceptual)

1. **People & surfaces** — Chat, Inbox, briefings, mindmap — where humans work.
2. **Context layer (Neither’s core)** — Structured organizational knowledge: knowledge graph (projects, people, decisions, dependencies, evidence) and APIs to query it.
3. **Agent mesh** — Coordination between agents (sessions, tools, audit).
4. **Agents** — Task executors that consume context via APIs and MCP-style tools.

Neither focuses on building the **context layer**: the graph and the interfaces that make it usable.

## Knowledge graph

Neither ingests email, documents, messages, and meeting content, then maintains:

- Projects — goals, milestones, status, health
- People — roles, organizations, communication patterns
- Decisions — provenance and evidence
- Dependencies — cross-project relationships
- Evidence — citations to sources

## Context API (directional)

The graph is exposed for programmatic access (authenticated, rate-limited, audited). Capabilities include workspace and project summaries, semantic search, evidence chains, and timelines — so partner systems and agent runtimes can query the same intelligence as the UI.

## Security & deployment

Production systems use encrypted transport and storage, workspace isolation, and hosted infrastructure suitable for business data. See **Security** for plain-language detail.

## Links

- Story: https://www.neither.online/story.md
- How it works: https://www.neither.online/how-it-works.md
- Security: https://www.neither.online/security.md