I have embedded ALL of Science and made it available for free, so your agents can now do semantic search over the whole scientific litterature.

I think this is big. I hope to provide casual use for free.

Point your agent at: https://sciencemadereadable.com/?api they will know what to do.

Friends don’t let friend’s agent use blogpost to ground their answers.

Of course, it also all opensource, links to repo on the page.

If you want to hit up the service at a higher rate then 10 req/min DM me.

the embeddings are powered be: lightonai/DenseOn

ScienceMadeReadable

Source: https://sciencemadereadable.com/?api

The library

Every paper already made readable. They open instantly.

Search science from your code and AI agents

The search behind this site is open to your programs and AI agents: about 200 million studies, by meaning or by exact words, with their citations and their place on the map of science. Free, with no key and no sign-up.

Full guide for agentsTry every callOpenAPI specSource code

Give it to your AI agent

The quickest way: tell it

Paste this into Claude Code, Codex, Cursor, or any agent that can make web requests:

When you need scientific papers, use the free search API at https://api.sciencemadereadable.com/api/data (no key needed). First read https://api.sciencemadereadable.com/api/data/llms.txt: it explains every call and how to search well.

Or install it as a skill

The agent then knows about the search in every conversation, and reaches for it by itself when research would help. One command, in a terminal:

Claude Code

mkdir -p ~/.claude/skills/science-search && curl -fsSL https://sciencemadereadable.com/agents/science-search/SKILL.md -o ~/.claude/skills/science-search/SKILL.md

Codex, Pi and other agents that read~/\.agents/skills

mkdir -p ~/.agents/skills/science-search && curl -fsSL https://sciencemadereadable.com/agents/science-search/SKILL.md -o ~/.agents/skills/science-search/SKILL.md

It’s a short text file:read itbefore you install it. Start a new conversation afterwards.

In a chat that can only open links

ChatGPT, Claude, Gemini and other chats that browse but can’t run code can search with a plain address: the question goes in the address, the answer comes back as text. Paste this:

To find scientific studies, open https://ask.sciencemadereadable.com/ followed by the question, for example https://ask.sciencemadereadable.com/why%20do%20cats%20purr (add ?k=30 for more results, ?year_from=2020 for recent ones). Cite the links it gives.

Try it yourself:ask.sciencemadereadable.com/why do cats purr. Some chats only open addresses that appear in the conversation: then paste the full link with your question in it.

No MCP server yet. The ways above work with any agent that can fetch a web address.

Call it from code

Plain HTTPS and JSON. Click a link to see a real answer in your browser.

B=https://api.sciencemadereadable.com/api/data
curl -G "$B/search" --data-urlencode "q=does intermittent fasting help weight loss" -d k=5 -d abstracts=false
curl -G "$B/search" --data-urlencode 'q="gut microbiome" AND (autism OR ASD) NOT mice' -d k=5
curl "$B/works/10.1126/science.aan8048"

import requests
B = "https://api.sciencemadereadable.com/api/data"
r = requests.get(f"{B}/search", params={"q": "microplastics in human placenta", "k": 20,
                                        "year_from": 2020, "abstracts": "false"}, timeout=30)
r.raise_for_status()
for p in r.json()["results"]:
    print(p["year"], p["citations"], p["title"], p["doi"] or p["openalex"])

/search?q=…Studies by meaning (ask in plain words) or by exact words (quotes, AND, OR, NOT). Filters for years, type, open access.Try ↗/works/\{id\}One study’s full record, by DOI, OpenAlex id or PubMed Central id.Try ↗/similar/\{id\}The studies closest in meaning to one study.Try ↗/walk?id=…A citation walk: what studies like this one cite that it does not. Good for “what should I read next?”.Try ↗/map/placeWhere studies, or a question, sit on the map of science.To send a person to a study on this site, link tohttps://sciencemadereadable\.com/?work=followed by its DOI.

Good to know

• **Be gentle:**30 requests a minute per address (reading the guide doesn’t count). Ask for up to 100 results in one search rather than many small ones. Go over and the answer says how many seconds to wait.
• **What it covers:**English titles and abstracts from OpenAlex, not full texts. New papers arrive with each data update, so the very latest may be missing. Theguidegives today’s exact numbers.
• Ask like a paper, not like a search box:“effects of dietary interventions on weight loss in adults” finds better studies than “how to lose weight”.
• **Judge results yourself:**each result has ajunkscore (likely spam or not research), but a clean score doesn’t mean a study is right. Subject labels are often wrong.
• **Privacy:**the server’s request log includes what was searched, but not who searched it.
• Study records come fromOpenAlex(CC0). The search code is open source (Apache 2.0).

Written byOriginal alongside

How it works

Scientific papers are written for other scientists. This page rewrites a whole paper for a curious 14-year-old: in the authors’ own voice, keeping every result, number and caveat, but in words anyone can follow.

The rewriting is done by small open models we trained(Qwen 3.5 with 9, 4 and 0.8 billion parameters). They learned from thousands of papers rewritten by frontier AI models. They are small enough to run on a gaming graphics card, and that is where they run: an RTX 3090 in my home. On busy days a rented Nvidia H100 joins it. When many people are reading, a paper may wait in line, and the page says when the GPU is offline.

**How good is it?**On our test papers, a judge model scored our 9B 7.8 out of 10, a little above GPT-6 Luna (7.5), below Claude Opus (8.5); it costs about half a cent per paper. It still makes mistakes more often than the big models, so check the original before relying on a detail. With your own API key you can compare with six commercial models.

**What can I search?**About 200 million studies: every English title and abstract in OpenAlex, the open catalogue of the world’s research. Search by meaning (we read your question like the studies) or by exact words. Each study sits on the map of science, near the studies most like it. Your own programs and AI agents can use the same search:here’s how.

**Which can be made readable?**Open-access research articles in PubMed Central under the CC BY licence, which allows republishing an adapted version with credit (about 5 million). Figures and tables are always the original ones. For the others we show what is known and link to the original.

A demo, not a product. We count visits on our own server, without cookies: each visitor is an anonymous code that changes every day, and the counts don’t include what you search for. Your search words do pass through our server’s technical log, without your address. If our search is down, “Readable only” searches go to Europe PMC instead, straight from your browser. If your browser asks not to be tracked, nothing is counted. Rewrites made by our models are kept so the next reader gets them instantly.