Agents

AI agents are often the easiest way to use OpenAlex — and frequently the most powerful. You describe what you want; the agent figures out the queries, calls the API, and hands you results. Agents already know OpenAlex well: for most tasks, “use OpenAlex” is all the setup they need.

Using the OpenAlex API, find the 25 most-cited papers about microplastics
published since 2024 and save them as a CSV with title, year, citations, and DOI.

This page is about how to work with agents effectively — which kind to use, how to set them up, and how far you can push them.

Which agent, for what

  • A chat agent (Claude, ChatGPT, Gemini) is right for one-off questions and small exports: “who are the most-cited authors at my university this decade?”, “get me these 200 DOIs as a spreadsheet.” Zero setup.
  • A coding agent (Claude Code, Cursor, Codex) is right when the output is a dataset or analysis: it writes real scripts that page through results, retries failures, and save clean files — and the scripts are yours to re-run and audit. This is the sweet spot for systematic reviews, bibliometric analyses, and anything you’ll want to reproduce later.
  • For big pulls, a coding agent can drive the CLI (“download all works on this topic as JSONL”) or work against the snapshot — you get bulk-scale results without learning the tooling yourself.

Set your agent up for success

Give it your API key. Agents make a lot of requests, and the keyless budget runs out fast. Sign up free at openalex.org, copy your key from Settings → API key, and paste it into the chat. (If a key ever leaks, rotate it in Settings — the old one dies instantly.) See Authentication for how budgets work.

Point it at the docs. Anything the agent is unsure of, it can read right here — this whole site is optimized for AI use. If it seems lost, saying “check help.openalex.org” is usually enough. For heavy API work, hand it the LLM quick reference, a condensed page written specifically for agents.

For agents that write queries programmatically, point them at OQO — the JSON query format with a schema to validate against, much harder to get wrong than assembling query strings.

Make it durable. If you use a coding agent regularly, tell it once — in its memory or config file — that you use OpenAlex and where your key lives. Every future session starts already set up.

Trust, but verify

Agents very rarely hallucinate OpenAlex results — the API is well-structured, and answers come from real responses, not the model’s memory. Still, for consequential work:

  • Ask to see the API calls it made. They’re URLs; you can open them yourself.
  • Spot-check a few rows against the website.
  • Prefer scripts over vibes for anything you’ll cite: a coding agent’s script is checkable and re-runnable in a way a chat transcript isn’t.
  • Quickstart — the five-minute version: website → API → agent
  • LLM quick reference — the condensed API reference to hand your agent
  • CLI — the command-line tool agents can drive for bulk downloads
View as Markdown