scime atlas · for developers and AI assistants
Use the atlas from code or an AI assistant
Send a research idea in one or two sentences and get back the closest accepted papers (ICML, NeurIPS, ICLR, CVPR, AAAI, 2016–2026), how crowded the area is, recent arXiv preprints and fast-growing approaches nearby. Free, no sign-in, no key.
Paste a result into ChatGPT, Claude or Gemini
After you place an idea on the map, press Copy for AI under the result. It copies the idea, where it lands, the nearest papers with links and the yearly counts as plain text, ready to paste into any AI chat with your own question.
REST API
GET https://atlas.scime.ai/api/v1/related?idea=<your idea>| Parameter | Meaning |
|---|---|
idea | The idea in English, 8–2,000 characters. Naming the method, the task and the data works best. |
k | How many nearest accepted papers to return, 1–20 (default 10). |
preprints | 0 leaves out recent arXiv preprints (default: included). |
format | text returns a plain-text summary for language models; default is JSON. |
Example
curl "https://atlas.scime.ai/api/v1/related?idea=speculative%20decoding%20with%20a%20small%20draft%20model&k=5"import requests
r = requests.get("https://atlas.scime.ai/api/v1/related",
params={"idea": "speculative decoding with a small draft model", "k": 5})
for p in r.json()["papers"]:
print(round(p["sim"], 2), p["t"], p["v"], p["yr"])The JSON has papers (title t, venue v, year yr, link s, similarity sim), clusters (area, topic, approach), closelyRelated and perYear (closely related papers per year, 2016–2026), preprints and openings.
Find a field
Look up an area, topic, keyword cluster or approach by name or keyword (abbreviations such as LLM, RLHF or NeRF work too). Each result has a path you can open in the Flow: https://atlas.scime.ai/?view=flow&path=0.1. Or open https://atlas.scime.ai/?q=continual+learning to jump straight to the best match.
GET https://atlas.scime.ai/api/v1/fields?q=continual%20learning&k=8MCP for AI assistants
Assistants that support the Model Context Protocol can add the atlas as a remote server at https://atlas.scime.ai/mcp. It offers one tool, find_related_papers, with the same inputs and results as the REST API.
Limits and terms
- 30 requests a minute and 100 searches a day per IP address. Write to contact@latenticlabs.com if you need more.
- The idea text is embedded on our server to find related papers and is not stored. Only an anonymous count of API calls is kept. See the privacy policy.
- Similarity is a guide, not a judgement of novelty. Preprints are not peer reviewed. Paper lists come from public accepted-paper lists and the arXiv API.