Build OpenAI assistants

The OpenAI Assistants API provides managed agents with persistent threads, built-in memory, and bundled tools: code interpreter for data analysis, file search for RAG over uploaded documents, and function calling for external systems.

It is the low-configuration middle ground: less bespoke than Claude, but with managed state and RAG that Glean-style search doesn’t cover when the data lives in uploaded files rather than an indexed corpus.

AESOP runs the generic pipeline — fitness check, gap analysis, workflow design, instruction writing — and emits the assistant’s configuration:

• Assistant config — model, instructions, and thread setup.
• Tools — function-calling definitions and code-interpreter enablement.
• Vector store — the uploaded files the assistant retrieves over.

The result is a single managed assistant the team can wire into their product, without standing up their own state management.

Deployment publishes the assistant to the OpenAI account. The deployed assistant is then run through AESOP’s Evaluate stage and scored for quality against the PRD.