---
title: "Blueprints — Context Infrastructure"
description: "Five architectures cover most of what teams build."
canonical: "https://press.getalchemystai.com/books/context-infrastructure-build-ai-that-understands-your-business/blueprints"
last_updated: "2018-10-20"
book: "Context Infrastructure"
label: "Chapter 12"
---

# Blueprints

Five architectures cover most of what teams build on Alchemyst. The sidecar pattern below is the one behind every industry case study in Part II.

**The context layer sits beside your systems and feeds every call.**

| Step | From | To |
| --- | --- | --- |
| ① Push | Systems of record (HIMS, CRM, DMS, ATS — nothing is replaced) | Context layer (one profile per customer, filtered retrieval) |
| ② Retrieve | Context layer | Voice agent (Kathan or your own, native Indian languages) |
| ③ Write back | Voice agent | Systems of record |
| ④ Escalate | Voice agent | Human team (the right named person) |

*Figure 12.1 · The sidecar pattern behind every voice case study*

1. **Push.** Each system of record sends context events to the API on key state changes: an appointment booked, a shipment flagged, a lead updated. Existing databases stay untouched.
2. **Retrieve before the call.** The customer's record, language and history are fetched and filtered before the call connects (170 ms at p50 in production voice deployments), so the agent opens with specifics.
3. **Write back.** Structured outcomes (bookings, summaries, objections, promises) flow back to the HIMS, CRM, ATS or DMS without a human touching them.
4. **Escalate with context.** Anything sensitive reaches the right named person with the full transcript and context attached.

## Five patterns

| Pattern | Sources | Scope design | What the agent does |
| --- | --- | --- | --- |
| **Support knowledge base** | Product docs, FAQs, resolved tickets | `[support, product \| faq \| tickets, …]` | Answers with source citations; every interaction is logged back as future context |
| **Engineering code assistant** | Repositories, design docs, API specs, PRs | `[codebase, src \| docs \| tests, …]` | Explains and suggests code inside the IDE, with links to the files and PRs it used |
| **Summarisation agent** | Reports, meeting notes, emails | `[content, department, period]` | Produces executive summaries with source attribution and flags contradictions |
| **Voice sidecar** | HIMS, CRM, DMS, ATS events | One profile per customer, scoped per campaign | Opens every call with context, remembers across attempts, writes outcomes back |
| **Cross-product identity** | Every product a customer touches | One canonical identity keyed on verified identifiers | Enriches existing agents at inference time through the Context Router |

**Three integration verbs.** In the real estate deployment pattern each product connects through one of three modes: **Push** (the product sends updates to the layer), **Pull** (the layer reads from the product) or **Route** (the Context Router sits in front of the product's LLM calls and injects context). Most estates mix all three, and none of them requires migrating the incumbent system.

**Targets worth copying.** The cross-product identity design aims for identity resolution covering more than 85% of active customers and end-to-end context lag under five minutes, with an audit trail on every read and write.

---

- Previous: [Chapter 11: Plug In](https://press.getalchemystai.com/books/context-infrastructure-build-ai-that-understands-your-business/plug-in.md)
- Contents: [Context Infrastructure](https://press.getalchemystai.com/books/context-infrastructure-build-ai-that-understands-your-business.md)
- Next: [Chapter 13: Read the Trace](https://press.getalchemystai.com/books/context-infrastructure-build-ai-that-understands-your-business/read-the-trace.md)
