On this page
What Asta does
Ai2 describes Asta as an agentic research tool with Find Papers, Generate a Report and Analyze Data. A report can use the open-weight AstaBrief Fast mode or Claude-powered Thinking mode; separate experimental tools include AutoDiscovery and Generate Theories. The official FAQ says the three main tools do not consume Asta credits, while some experiments do. This source-based profile explains those documented boundaries and the FAQ’s data-handling statements alongside Ai2’s general terms. No account session, paper search, cited report, theory run or dataset analysis was performed.
Choose an Asta research task
Decide whether you need a candidate paper set, a cited overview, a data exploration or an experimental hypothesis workflow.
| Your task | Start with | Check before continuing |
|---|---|---|
| Find papers for a research question | A scoped question, search boundary and one expected source to check | Open candidate papers and record relevant results the search did not return. [1] |
| Draft a cited overview of literature | A defined question and papers you can inspect in full | Follow every citation to its source and check how the report represents methods and limitations. [1] [2] [7] [8] |
| Explore an approved research dataset | A non-sensitive structured file with known columns, units and missing-value rules | Compare the output and chart with the source data; do not upload regulated material to hosted Asta. [2] |
| Generate candidate theories from papers | A bounded question and awareness that the retrievable full-text set may be incomplete | Read the cited passages and scope; “novel” is relative to the retrieved set, not all research. [2] |
Where it fits, and where it falls short
Useful strengths
- Find Papers decomposes a research question into components, follows citation links and provides relevance explanations for candidate papers.
- Generate a Report searches Ai2’s abstract and full-text indexes to assemble a cited mini-review; Ai2 describes it as distinct from generating candidate theories.
- Generate a Report offers the open-weight AstaBrief Fast mode alongside Claude-powered Thinking mode, according to Ai2’s October 2026 release notes.
- Analyze Data accepts several structured research-data formats and can produce summaries and visualizations from natural-language requests.
- The official FAQ separates the credit use of core tools from experimental workflows and describes data and interaction controls for Analyze Data.
Limits to consider
- Asta’s search and reports depend on the indexed corpus and retrievable full text. A missing paper or paywalled source is a coverage gap to investigate, not evidence that the wider literature contains no relevant work.
- AstaBrief’s model card describes an input of a research question and already-retrieved literature excerpts. The published weights are a model component, not the complete Asta search interface or retrieval workflow; local use requires a compatible inference setup and a way to supply sources.
- Generate Theories is an experimental candidate-generation tool, not a literature review. Its “novel” label is relative to the papers retrieved for that run, and support/conflict counts represent findings rather than independent papers.
- The FAQ says hosted Asta must not receive sensitive or regulated data. Use a self-hosted deployment only where the institution has established and approved its own controls.
- Asta’s FAQ says uploaded datasets are not used for model training and interaction training requires explicit opt-in. It also says the FAQ is a high-level summary that does not replace accepted terms; Ai2’s general Terms of Use reserve broader rights to use User Content unless specified otherwise. Review the terms and consent controls presented for the specific service.
- Credits are an internal compute measure, not an accuracy signal or a charge by themselves. A credit-free Find Papers or report run can still omit, misread or misattribute evidence.
- This profile summarizes public documentation only. It does not report an account trial, paper-recall benchmark, citation audit, data-analysis result or independent privacy review.
A suggested Asta workflow
View the workflow diagram
- 1
Write the question, population or subject, date boundary and evidence type you need before searching. Add one known relevant paper or a plausible exclusion as a check on what the result set includes.
- 2
Use Find Papers for candidate discovery, then open the original papers. Record which results are abstracts, which full texts you can inspect and which expected sources are absent. A relevance explanation is a lead, not a screening decision.
- 3
Use Generate a Report for a cited synthesis only after deciding which papers belong in the answer. Follow each citation to the original passage, compare the claim with its methods and limitations, and retain the paper locator in your notes.
- 4
For Analyze Data, use a copy of an approved, non-sensitive dataset. State the outcome, units and comparison; inspect column meanings, missing values, filters and any generated code or visualization against the source file before interpreting it.
- 5
If you try AutoDiscovery or Generate Theories, preserve the retrieved set and experiment settings. Treat a proposed pattern as a candidate to verify in the cited papers, not as a new result or proof of research priority.
Separate paper discovery from a literature report
Ai2 describes Find Papers as a query-led search that breaks a question into components, follows citations, assesses relevance and explains why a candidate paper matches. Use those explanations to decide what to open, not to stand in for your own inclusion criteria. Record the question and date boundary so you can inspect what was searched and what was missed.
Generate a Report is a different task: Ai2 says it searches its abstract and full-text indexes, re-ranks passages and produces a cited mini-review. Keep the report’s sources visible. Open the papers, verify each supporting passage and note whether the wording describes an abstract, a full-text result or your own interpretation. A citation list does not establish that every included study is relevant or that the set is comprehensive.
Ai2’s October 2026 release article says Generate a Report includes Fast mode, powered by the open-weight AstaBrief 8B model, alongside Claude-powered Thinking mode. The AstaBrief model card describes its input as a research question plus retrieved scientific literature excerpts. This makes the model a report-generation component rather than a paper-search interface by itself. The model card lists an Apache 2.0 license; a local setup still needs an inference environment and an appropriate source-retrieval workflow. Open weights alone do not certify an institution’s privacy or data controls.
Avoid treating Generate Theories as another summary mode. Ai2 explicitly describes it as proposing testable candidate claims across retrieved papers and recommends Generate Report for a literature review. Compare the task and source set before choosing a tool.
Check the dataset, permitted use and analysis assumptions
The Asta FAQ lists CSV, Excel, JSON or JSONL, HDF5, TSV and Parquet for structured data. Before asking a question, document the file version, column definitions, units, missing-value codes and any transformations already applied. Ask a bounded question that can be answered from those fields, then compare the returned summary or visualization with the original rows and a calculation you can reproduce.
The FAQ says not to upload sensitive or regulated data to Ai2’s hosted deployment; it identifies self-hosting as the route for data subject to requirements such as HIPAA or GDPR. This is a provider instruction, not a determination that a particular institution’s self-hosted setup meets its obligations. Follow your institution’s approval, data-use agreement and deployment controls.
For hosted Analyze Data, Ai2 says uploaded source files are automatically purged from its systems seven days after upload unless deleted earlier. For signed-in users, summaries and results remain accessible after the source files are purged; deleting the exploration removes its datasets, analyses, results and outputs immediately and permanently. Explorations become read-only after 24 hours. Anonymous explorations with no activity are deleted after 30 days. These retention statements apply to Analyze Data as described in the FAQ, not necessarily to every Asta prompt or feature.
Treat generated theories as candidates, not conclusions
Ai2 says Generate Theories finds up to 100 relevant papers and proposes patterns with scope, predictions and evidence. The FAQ says it only sees open-access full text it can retrieve, may miss papers and cannot see paywalled work or gray literature. A result can therefore be interesting within the retrieved set while still missing important contrary work.
The FAQ distinguishes a literature review from a theory run: the latter proposes candidate claims rather than summarizing what papers say. It says “novel” means relative to the papers retrieved in that run, not all science, and that supporting or conflicting totals count findings rather than papers. Inspect the evidence tab and each cited article; do not describe these counts as a systematic review or established discovery.
Ai2 reports that a theory run usually takes 20–30 minutes and may use up to 100 credits depending on the papers used. The FAQ also says failed runs return unused credits. Treat these as current provider descriptions and confirm the credit estimate before starting a substantial run.
Supporting documentation: [2]
Understand credits and experimental feature boundaries
Ai2 says a new account receives a one-time grant of 1,000 Asta credits. The FAQ lists Find Papers, Generate a Report and Analyze Data as credit-free; AutoDiscovery costs one credit per experiment, while Generate Theories uses one credit per source paper. If your balance drops below 120, the FAQ says you may request 100 additional credits. Credits track compute use and are not a price-per-paper estimate.
The Asta FAQ calls Preview an early-access program for evolving features and prototypes, and says participants are expected to try them and give feedback. Confirm whether a tool is part of the standard hosted product or the Preview program before planning a course, lab or institutional workflow around it. The profile does not claim all tools have the same access status.
Supporting documentation: [2]
Read the Asta FAQ together with Ai2’s accepted terms
For Analyze Data, the Asta FAQ says uploaded datasets are not used for model training, sharing or redistribution, and interactions are not used for model training without explicit opt-in. The FAQ calls itself a high-level summary that does not replace accepted terms and points readers to Ai2’s Privacy Policy for its full privacy practices. Ai2’s general Terms of Use say that, unless a service specifies otherwise, User Content may be used to train models and grant Ai2 broad rights to use content. Read the terms and any service-specific notice or consent control shown when you sign in; do not treat the FAQ as a blanket guarantee for every prompt, file and output.
The FAQ says limited, least-privilege administrator access to uploaded datasets may occur for safety, reliability or legal reasons, and that users are responsible for confirming they may upload the material. Avoid confidential, unpublished, identifying or restricted records on the hosted service unless the responsible institution has approved the exact service, terms, source material and retention conditions.
A suggested exercise: verify a report claim before using it
Choose a narrow question from a public, non-sensitive research topic and write down one paper you expect to find. This is a proposed exercise, not an Asta session or a result we measured.
Run a paper search and record the candidate sources, then compare them with the paper you expected. Select one claim from a report and open the cited article. Check whether the passage supports the same population, outcome, method and degree of certainty. If the source is an abstract, say so; do not imply that you checked unavailable full text.
Write one sentence about what the evidence supports and one about what remains unknown. If the tool’s search missed a known source, keep that as a coverage gap to investigate rather than concluding that the source is irrelevant. Do not use a generated “novel” label to claim priority.
Pricing, free access and plan checks
Ai2’s Asta FAQ describes credits as a usage measure rather than a payment and says a new signup receives 1,000 credits once. Find Papers, Generate a Report and Analyze Data are listed as using no credits; AutoDiscovery uses one credit per experiment, and Generate Theories uses one per source paper. When a balance falls below 120 credits, the FAQ says the user can request 100 more. These figures describe the published credit system, not a guarantee of service availability, subscription price or unlimited access. Check the current sign-up flow and accepted terms before use.
Confirm billing period, renewal terms, limits and regional availability on the provider’s current plan page before paying.
Questions about Asta
What is Asta AI?
Ai2 describes Asta as a scientific research tool with paper discovery, literature-report generation and structured-data analysis, alongside experimental discovery and theory tools. It is not a generic essay-writing service.
Does Asta search for research papers?
Its Find Papers tool decomposes a research question, follows citations and explains relevance for candidate papers. Search results still need screening, and a missing paper should be investigated rather than treated as proof of no relevant work.
Can Asta write a literature review?
Ai2 describes Generate a Report as a cited mini-review based on its indexed abstracts and full texts. Generate Theories is separate: it proposes candidate claims across retrieved papers and is not a literature review. Verify the report citations and source coverage yourself.
Can I run AstaBrief locally?
Ai2 publishes open AstaBrief 8B weights and a model card that lists an Apache 2.0 license. The model takes a research question and retrieved literature excerpts to produce a cited report; it is not the complete Asta search interface. Local use requires an inference setup and a way to retrieve and provide source excerpts. Review the license and your institution’s data controls before using research material.
Is Asta free?
The Asta FAQ says credits are a usage measure and users are not charged for the credits themselves. It describes one-time signup credits and says Find Papers, Generate a Report and Analyze Data use no credits. Check current account access, feature status and accepted terms; the credit FAQ alone does not establish that every service is unlimited or universally available.
Can I upload a research dataset to Asta?
The FAQ lists common structured formats for Analyze Data but says not to upload sensitive or regulated data to hosted Asta. It identifies self-hosting as the route for regulated datasets. Confirm institutional approval and the service terms before using any research data.
Does Asta train models on my files or prompts?
The Asta FAQ says uploaded datasets are not used for model training and interactions are not used without explicit opt-in. The FAQ also says it does not replace accepted terms; Ai2’s general Terms of Use contain broader provisions unless a service-specific rule applies. Review the applicable terms and consent controls for the exact feature before submitting material.
Does Asta prove that a theory is novel?
No. Ai2 says “novel” is relative to the papers retrieved for that run; Generate Theories may miss papers and does not see paywalled or gray literature it cannot retrieve. Treat the result as a candidate and verify its evidence and scope.
Has AI Essay tested Asta?
No. This is a source-based profile of Ai2’s public documentation. We did not run an account, search, report, theory generation or dataset analysis, and we do not claim a citation-accuracy or research-quality benchmark.
Sources and research notes
This guide synthesizes official product information with our editorial assessment of academic use. Suggested exercises are examples you can repeat; they are not results of tests we performed. We do not infer accuracy from a provider’s marketing claims.
- Ai2: Asta agents and documented scientific workflows
- Asta FAQ: credits, data analysis, Preview and Generate Theories
- Ai2: Asta scientific research ecosystem and access route
- Asta hosted research workspace
- Ai2 Privacy Policy
- Ai2 Terms of Use: User Content, AI services and model training
- Ai2: Open-sourcing AstaBrief, the fast report-generation model in Asta
- Ai2 AstaBrief 8B model card and license
Found a changed feature or a factual error? Send the exact page and an official source to info@ai-essay.com. Read our editorial policy for how we handle corrections.