Knowledge as of Oct 26, 2026
The Inference TimesReference architectures, tutorials, demos and free skills — by Nitin Goswami
Epoch III · Step 16Ask the AI Nitin →

Section

Reference Architectures

Opinionated, production-tested designs you can adapt.

Reference Architecture · Agents

Agent memory over MCP: giving any model a long-term memory it can call

Agent memory over MCP puts an assistant's long-term memory in its own small server, so that any app that speaks the Model Context Protocol can save facts and look them up again. Explained with three diagrams and one worked example.

By Nitin Goswami · · 8 min read

A language model forgets everything when a conversation ends. Agent memory over MCP fixes that by running a memory server: a small program that stores memories in a vector store and offers them as tools the model can call, over the Model Context Protocol (MCP), an open standard for connecting AI apps to outside tools and data (MCP specification, 2025). In the terms of the CoALA framework (Sumers et al., 2023), saving…

Continue reading →

Reference Architecture · Voice AI

Grounded voice insight: turning recorded speech into claims you can check

A pattern for getting judgements out of recorded conversations that a reader can verify: transcribe, measure what can be counted, let an LLM judge the rest with quotes, and check every quote against the transcript before showing it. Explained with two diagrams.

· 5 min read

Reference Architecture · Agents

What is an internal agent SDK? One shared way for many teams to build AI agents

A shared toolkit that lets every team in a company build AI agents the same safe way: each team writes only what is unique to its agent, and the SDK supplies the checks, the step-by-step runner, saved progress, human sign-off, model access and cost limits. Explained with two diagrams.

· 9 min read

Reference Architecture · Agents

CoALA: how language agents remember, decide and act

A reference architecture for thinking about any LLM agent: three kinds of long-term memory, a working memory where the model reasons, an action space, and a decision cycle that ties them together. Explained with two diagrams and one worked example.

· 6 min read