Frequently asked questions
What is AstroFetch?
A conversational interface for astronomical data archives. You ask a question in your own words; AstroFetch translates it into ADQL, runs the query against the archive's TAP service, and returns results plus analysis. It has built-in tools for plotting, Python execution on result sets, and NASA ADS literature search.
What archives does it support today?
The NASA Exoplanet Archive: confirmed planets, planet candidates, and stellar properties.
Additional TAP archives are planned for future releases. If there's an archive you'd like to see supported, send astrofetch-help@ipac.caltech.edu a note.
Can I bring my own API key?
Yes. Click the key icon in the sidebar, paste your Anthropic API key, and choose Anthropic as the provider. BYOK requests bypass the institutional rate limit and use your own provider billing. During the beta evaluation period, only Anthropic keys are accepted; other providers will be re-enabled after eval.
Why did I get blocked from sending a prompt?
During the beta evaluation period, you must rate the most recent response before sending another prompt; these ratings are used to improve the system. Click any star on the previous response to unblock the input. Comments are optional but appreciated.
How accurate are the queries?
Generally good for well-formed questions, but treat results as drafts until you've checked them.
AstroFetch displays the ADQL it ran with each query; review it before relying on the result. Wrong column names and unit mismatches are the most common error types. The agent has access to curated column documentation but occasionally picks the wrong column when there are similarly-named ones (e.g., pl_rade vs. pl_radj for planet radius).
For anything you'd cite or publish, verify the result against the archive's own documentation and re-run the query through your usual workflow as a sanity check.
Can I download query results?
Yes. Each query produces a sortable, scrollable table with a CSV download link. Use the CSV as the canonical artifact for anything you'll cite.
Plots produced by the chat are illustrative only - for publication-quality plots, download the CSV and produce them in your own workflow.
What happens to my chats?
Stored per-user in our PostgreSQL database on AWS (IPAC's account). You can see all your chats in the sidebar, delete individual ones, or request full account deletion via astrofetch-help@ipac.caltech.edu.
During the beta evaluation period, AstroFetch admins (currently the NExScI team) can view all chats for quality review and debugging. We do not share your chats with anyone outside the project.
How do I report a bad response?
Click 1 or 2 stars on the response, optionally add a comment in the popup, and submit. Comments are especially valuable when something looks wrong - even a brief note such as “wrong column” helps us prioritize.
We review collected feedback at the end of the beta evaluation period and use the patterns to refine the system prompt and archive profile docs.
Is my data sent to Anthropic, OpenAI, or Google?
By default, every prompt is sent to Anthropic's API along with your chat history and the relevant archive's domain context.
Because we use a paid Anthropic API key rather than the consumer Claude product, Anthropic's commercial-API terms apply: your prompts and the model's outputs are not used to train Anthropic's models. See Anthropic's commercial terms for the authoritative policy.
AstroFetch does not share your data with anyone outside the project. If you bring your own API key, requests route through your provider account and your provider's data policy applies.
What model is the agent using?
Claude Opus 4.7. During the beta evaluation period, Opus is the only selectable model. Additional models and providers will be considered after the beta evaluation period concludes.
How is AstroFetch different from just using ChatGPT or Claude directly?
What sets AstroFetch apart isn't the underlying model - the same Claude or GPT family is available in any chatbot. The difference is the curated, archive-specific context wrapped around it.
For each TAP archive we support, we maintain hand-written system prompts and agent-facing column-level documentation: which tables answer which kinds of questions, the canonical units, common pitfalls (e.g., pl_rade vs. pl_radj for planet radii), and example queries that demonstrate idiomatic ADQL for the archive's schema. The agent is grounded in the same documentation an experienced archive user would consult.
A general-purpose chatbot has access to the public web but not to that curated knowledge base. It tends to invent column names, miss the right table, or guess at units - problems that compound when it can't actually run the query and verify the result. AstroFetch can.
What's the rate limit?
The institutional Anthropic key has a daily per-user prompt cap. When you reach the cap, the app shows a banner with two options: wait until the daily reset, or supply your own Anthropic API key to continue. BYOK requests have no AstroFetch-side rate limit.
Who runs this?
NASA Exoplanet Science Institute (NExScI) / Caltech-IPAC.
For questions, bug reports, or feedback that doesn't fit a star rating: astrofetch-help@ipac.caltech.edu
How should I acknowledge AstroFetch in publications?
If results from AstroFetch contributed to your paper, please include the following acknowledgment text:
This research has made use of AstroFetch, an AI-powered conversational interface for astronomical data archives developed by the NASA Exoplanet Science Institute (NExScI) / Caltech-IPAC.
In addition, please include the citation guidance from each archive whose data you used. AstroFetch is a conduit to those archives, not a substitute for them - the underlying data still needs to be credited per the archive's own policy. The link to the currently-selected archive's homepage is in the second line of the chat disclaimer.