Blog

Notes on better support

Practical guides on AI support, grounding answers in your knowledge, and getting the most out of AskBrew.

Engineering · 8 min read

API-based AI support: integrating an assistant into your own product

When a hosted widget does not fit — custom interfaces, mobile apps, existing helpdesks — and what an integration needs to get right.

Read article →

Comparison · 7 min read

AskBrew vs Chatbase: which fits your support team?

Both build a chatbot from your own content. They differ in what happens when the assistant is unsure, and in whether it can answer questions about a specific customer's account.

Read article →

Buyer's guide · 8 min read

Chatbase alternatives: how to actually choose one

A buyer's guide to the AI support chatbot category — the four questions that separate these tools, and which type of alternative fits which situation.

Read article →

Buyer's guide · 9 min read

Best AI customer support software: a buyer's guide for 2026

What AI support tools actually do, the three categories they fall into, and how to evaluate them without relying on vendor feature tables.

Read article →

Engineering · 9 min read

Multi-tenant RAG: keeping one customer's data out of another's answers

Vector search across shared infrastructure has a failure mode worse than a bad answer. How to make tenant isolation a property of the database rather than a discipline.

Read article →

Engineering · 9 min read

Function calling for support bots

Letting an assistant look up a customer's order turns a content bot into something useful — and introduces an authorisation problem that must be solved outside the model.

Read article →

Engineering · 12 min read

How to build an AI support agent

A build guide for the whole system: ingestion, retrieval, the answer gate, tool calls, and the operational pieces tutorials leave out.

Read article →

Engineering · 9 min read

Preventing hallucinations in customer support bots

Grounding reduces fabrication but does not eliminate it. The techniques that actually work, and why an assistant that refuses is more valuable than one that always answers.

Read article →

Engineering · 8 min read

pgvector vs Pinecone: choosing a vector store

A dedicated vector database and a Postgres extension solve the same problem differently. The deciding factor is usually not search performance.

Read article →

Engineering · 10 min read

RAG architecture explained, for people shipping it

The components of a retrieval-augmented generation system, the decisions that actually affect answer quality, and where production implementations diverge from tutorials.

Read article →

Engineering · 8 min read

How AI customer support actually works

What happens between a customer typing a question and an answer appearing: retrieval, grounding, confidence, and the decision to answer or hand over.

Read article →