information-retrieval

An LLM-written digest of the most notable material definitive agreements filings in the window.

An LLM-written digest of the most consequential DevOps / cloud-native infrastructure releases in the window.

An LLM-written digest of what the tech community is focused on right now.

Fetches one public policy page in a real browser and turns it into a fixed schema: return window in days and what it counts from, who pays the return shipping, how the refund is issued and within how many days, any restocking fee, the required condition of the goods, the excluded product categories, the steps of the procedure, whether an RMA is needed, the contact route and whether a statutory right of withdrawal is mentioned.

Search CoinCap assets by name or symbol (agent-optimized).

Reads a Snapshot space and returns what a voter or an observer needs to act. Per proposal: the choices with their scores and share, total voting power cast, the leading option, the number of voters, the quorum and whether it is met, when voting closes and how much time is left. Scores are marked final or pending — Snapshot computes them asynchronously, and a pending score can still move, so a result read too early is not a result.

Queries OSV.dev for a list of packages with versions across npm, PyPI, Go, Maven, crates.io, NuGet, RubyGems, Packagist and the Linux distributions. For each package it returns the vulnerabilities found with their identifiers, severity and summary, and the single figure that decides what to do next: the lowest version that fixes all of them, derived from the fixed-version events in the affected ranges.

Fill a caller supplied JSON Schema from one web page and report source and confidence for every field.

Searches OpenAlex and Crossref for publications and returns each with its DOI, authors and their affiliations, publication year, venue, abstract, citation count and, where one exists, a direct link to a legally free full text with the licence it is offered under. OpenAlex stores abstracts as an inverted word index rather than text; it is reconstructed here, because the raw structure is unusable. Results can be filtered by year and restricted to open access.

Looks up a barcode number across Open Food Facts and a general product database, returning product name, brand, category, packaging, quantity and images where present. The check digit is computed locally before anything is queried: an invalid number is a typo, and reporting it as not found would send the caller looking in the wrong place.

Validates output against a JSON Schema and reports each violation with its JSON path, the rule it broke and a sentence saying what to change. Accepts either a parsed object or the raw string a model returned: a payload wrapped in a code fence or surrounded by prose is unwrapped, and the wrapping is reported separately from schema errors so you fix the right layer. Draft 2020-12 and draft-07 are both supported, along with the common string formats.

Takes two documents, by URL, inline base64 or raw text, extracts the text of each and compares them. Paragraphs are the unit: the answer lists every block that was removed, added, or kept but reworded, and for reworded blocks it names the individual words that went and came, so a single changed number in a long clause is visible instead of the whole clause being flagged. Alongside the content diff a structural comparison reports headings, paragraph, list and table-line counts on both sides.

Scores every catalogue model against one task and returns a ranked shortlist. You state the task kind, optionally a per-call budget in USD, how much speed matters, the input size and any vendor restriction; the answer names a model, the estimated cost per call, and the sub-scores it was chosen on, so the ranking can be checked rather than trusted. Models that cannot hold the input are excluded and listed separately with the reason. Weights are yours to set and are echoed back.

Viewport or full page, optional element selector, dark mode and cookie banner removal. Returns base64 image plus a short lived fetch URL, byte size, load time and the final HTTP status of the page.

Takes a file digest and answers which released package version it is. Useful for an artefact found in a container layer, a build directory or an attachment with no metadata attached. Hex and base64 input are both accepted and normalised, the digest length is checked against the declared algorithm before anything is fetched, and the response names every matching package version, since one artefact can be published under more than one coordinate.

Called without "fields" it inspects: every AcroForm field of the PDF comes back with its name, type, current value, whether it is required or read-only, the permitted options for dropdowns, radio groups and list boxes, and the character limit of text fields. Called with "fields" it fills exactly those and returns the resulting PDF as base64, together with what each field held before and after.

Retrieves water quality results for a monitoring site from the Water Quality Portal and normalises them into a comparable shape. The real work is units: the same measurement appears there as mg/L, ug/L, ppm and mg/l in mixed case, and combining them unchecked puts factor-of-thousand errors into any statistic built on top.

Takes any common identifier for a chemical and returns the canonical PubChem compound: the CID, molecular formula, molecular weight, canonical and isomeric SMILES, InChI and InChI key, the IUPAC name, and the computed properties that decide most downstream work — logP, exact mass, polar surface area, hydrogen bond donors and acceptors, rotatable bonds and formal charge.

Searches the device recall database for a product, manufacturer or model and returns each matching recall with the classification and its meaning, the reason, the recalling firm, the dates, the status and — the field that decides whether a specific unit is affected — the product description including lot numbers and serial ranges, quoted verbatim.

Retrieves occurrence records for a species within a radius of a coordinate and returns each with what makes it usable or not: the observation date, the coordinate uncertainty in metres, the basis of record — whether it is a preserved specimen, a human observation or a machine detection — the publishing dataset, and the licence of that record.

Sanctions screening (AML/KYC): screen a person or company name against 6 official watchlist sources — UN consolidated, EU FSF, US OFAC SDN, UK Sanctions List, French national asset-freeze register, Swiss SECO list — six official lists in one call. Fuzzy watchlist matching with a 0-100 confidence score per hit — never a bare yes/no. Each list reports its entry count, publication date and what that date means. One AML screening call covers UN, EU, OFAC, UK, FR and CH.

Takes a transcript as text, a public URL or inline base64 and returns what follows from the meeting: the tasks, who is responsible, by when, the decisions that were actually taken, and the questions that stayed open. Every single entry carries the verbatim passage it was drawn from, and that passage is looked up in the transcript before the answer goes out — an exact match or a match after whitespace is normalised.

Takes a contract as text, a public URL or inline base64 — PDF, DOCX, ODT, HTML and Markdown are read — and returns the passages governing the requested topics. Termination, liability, term, payment and jurisdiction are the default; confidentiality, data protection, IP, warranty, force majeure, assignment and notice can be requested as well.

Searches the clinical trial registry and returns matching studies with the fields that decide whether a trial is relevant: the recruitment status translated from the registry code, the phase, the study type, the enrolment target, the conditions and interventions, the sponsor, the start and completion dates, and the countries with sites.