Literature analysis for astrophysics questions
The literature agent answers a research question from the published record. It retrieves the relevant articles, reads what they claim, and returns a structured answer with a citation behind each interpretative statement.
What you get back
Rather than a list of isolated citations, the answer organises prior evidence, assumptions and disagreements: which studies agree, where they differ, and what is left unresolved. Each reference is given with its bibcode and linked to its NASA ADS record.
This context is then used by the other agents. Data analysis and model fits in the same session are compared against what has already been shown, contested or left open.
How well retrieval works
Retrieval was evaluated on two benchmarks over about 21,000 blazar papers. In the single-paper benchmark each of 209 questions targets one publication; in the multi-paper benchmark 154 questions may need evidence from several. At least one relevant paper was among the top five results for 76.6% and 79.2% of the questions respectively. The method and results are in the AstroGenesis paper, arXiv:2609.28579.
Example
Asked where gamma rays originate in the jet of 3C 279 during flares, the agent found that the literature does not place all flares at one distance: the brightest, fastest flares fit inside or near the outer edge of the broad-line region, while other major flares sit at sub-parsec to parsec scales. The answer cites eight papers. The current implementation focuses on blazar research.
Worked example
Other topics
- Blazar research with AstroGenesisThe first source class covered: literature, SED data and jet emission models in one session.
- Multi-wavelength and multi-messenger data retrievalScience-ready radio to gamma-ray data from MMDC, with instrument and epoch metadata.
- SED modeling with neural networks trained on radiative simulationsEmission models fitted to SEDs with neural networks trained on radiative simulations.
- Research ideation from literature, data and modelsOpen questions found where literature, data and models disagree.
Ask about your own source
Ask in plain language: literature, data, or a model fit. Free plan: 10 messages per day.