---
title: "The Layer — Context Infrastructure"
description: "One API that stores your knowledge and your agents' memory outside any model."
canonical: "https://press.getalchemystai.com/books/context-infrastructure-build-ai-that-understands-your-business/the-layer"
last_updated: "2018-10-20"
book: "Context Infrastructure"
label: "Chapter 2"
---

# The Layer

Alchemyst AI is a context layer: one API that stores your company's knowledge and your agents' memory outside any model, and hands each request only the context it should see. It is built by XAlchemystai Technologies Pvt. Ltd., India.

**Context lives in its own layer, between your systems and any model.** Sources flow in, scoped context flows out, outcomes flow back.

| Your sources | Alchemyst context layer | Any model, any agent |
| --- | --- | --- |
| **Knowledge** — docs, policies, FAQs | **Ingest** — bulk or async, deduplicated | **Agents** — support, ops, sales |
| **Systems of record** — CRM, HIMS, ATS, DMS | **Enrich and index** — institutional knowledge graph | **Voice** — Kathan Voice OS |
| **Interactions** — chats, calls, tickets | **Context arithmetic** — scope, combine, exclude, rank | **Developer tools** — IDEs and clients via MCP |
| **Engineering** — code, PRs, specs | **Context Traces** — sources, scores, rules applied | **Models** — any LLM, swappable |

Outcomes and new facts are written back to the layer.

*Figure 2.1 · Where the context layer sits*

Read it left to right. Your sources flow in once; every agent and model draws scoped context from the same layer; outcomes are written back so the next request starts from the latest truth. The highlighted step, context arithmetic, is what decides what each request sees.

## What it replaces

Teams that build context handling themselves usually assemble four pieces: a vector database, a graph database, a cache and a logger. Alchemyst provides all four behaviours behind one API, so there is no retrieval infrastructure to operate.

## What it is, and what it is not

| Alchemyst is | Alchemyst is not |
| --- | --- |
| A shared knowledge layer that many agents read from | A model, or a wrapper that locks you to one |
| Persistent memory that survives sessions and model changes | A replacement for your CRM, HIMS or data warehouse |
| A sidecar that sits underneath existing agents and telephony | A rip-and-replace platform |
| Auditable: every retrieval can be traced to its sources | A search box for human employees |

## Four properties to remember

- **Model-agnostic.** Context is stored separately from the LLM that uses it, so you can switch or route between models with zero context migration.
- **Context sovereignty.** The knowledge base is yours and portable; it does not live inside one vendor's assistant.
- **Fast.** Context retrieval runs under 300 ms at p95 across query types; Chapter 19 gives the test conditions.
- **Traceable.** Context Traces show which sources, scores and rules produced each context window.

---

- Previous: [Chapter 1: The Blind Spot](https://press.getalchemystai.com/books/context-infrastructure-build-ai-that-understands-your-business/the-blind-spot.md)
- Contents: [Context Infrastructure](https://press.getalchemystai.com/books/context-infrastructure-build-ai-that-understands-your-business.md)
- Next: [Chapter 3: Six Ideas](https://press.getalchemystai.com/books/context-infrastructure-build-ai-that-understands-your-business/six-ideas.md)
