GovQuery

Retrieve the full contents of a specific PIA document by its unique identifier (the `id` returned by `search`). This tool exists only to satisfy OpenAI's MCP search/fetch spec — use it ONLY alongside `search` when serving ChatGPT Deep Research or Company Knowledge. For all other queries, use `pia_search` / `pia_oversight_recommendations` instead.

fetch

Search the Program Integrity Alliance (PIA) GovQuery database of U.S. government oversight material (GAO, Inspectors General / Oversight.gov, CRS, DOJ, Congress.gov, Federal Register), returning relevant results with titles, snippets, and URLs for citation. This tool exists only to satisfy OpenAI's MCP search/fetch spec — use it ONLY when serving ChatGPT Deep Research or Company Knowledge, which require these exact tool signatures. For all other / general PIA queries, use `pia_search` instead (or `pia_oversight_recommendations` for open federal recommendations), which return richer, purpose-built results.

search

Search the Program Integrity Alliance (PIA) database of government oversight reports, recommendations, executive orders, legislation, and integrity data (GAO, OIG/Oversight.gov, CRS, DOJ, Congress.gov, Federal Register). One tool for all document search: scope by source/dataset/agency/date via filter, choose content vs titles via search_mode, sweep every source with wide, or discover filter values with facets_only. WHAT THE COUNTS MEAN: in 'content' mode (default) total_count and facet counts are TEXT CHUNKS / EXCERPTS — a single document is split into many chunks, so these counts are much larger than the number of documents and must NOT be reported as a document/report/article count. In 'titles' mode the counts are whole DOCUMENTS / ARTICLES. Use search_mode='titles' whenever the user asks how many documents, reports, or articles there are. Citations are very important to PIA users, so it is strongly expected that your answer: (1) gives every factual claim at least one clickable inline citation in the form [[1]](url), numbered sequentially from 1; (2) ends with a References section listing every source you cited, each reference on its own line (never multiple references on one line) — omit this section only for a pure count or summary with nothing specific to cite; and (3) includes a Find Out More section with the `govquery_url` so the user can open the full result set. Please do not present findings from these results without their inline citations and References section. Counts from free-text queries are approximate, because semantic search matches variations of the search terms.

pia_search

Search oversight recommendations (Open Recommendations dataset) with facets enabled by default. Today's date is 2026-10-21. This dataset contains recommendations from GAO and Oversight.gov only — a SourceDocumentDataSource filter naming either of those DOES narrow the count (use it for e.g. GAO-only questions); filters naming any other source are ignored. Do NOT filter by SourceDocumentDataSet — this tool already targets the 'Open Recommendations' dataset automatically. This tool reports the OPEN population only, so do NOT filter for closed recommendations (e.g. RecStatus eq 'Closed') — that always yields 0. SourceDocumentTitle, RecRecommendation and RecAgencyComments are NOT filterable here and are rejected with an error — put that wording in `query`, or use pia_search. IMPORTANT: `referenced_agencies` is a FILTER FIELD, never a parameter of this tool — passing it as an argument is an error. Express it only inside `filter`, e.g. filter="referenced_agencies/any(a: a eq 'Department of Homeland Security (DHS)')". When a text query is provided, do NOT add that agency filter at all — the query already handles relevance matching. Only add it when the query is empty (pure filter/facet lookups). `total_count` is the number of OPEN recommendations matching the query and all applied filters — report it as that count. The `results` are a representative sample to summarize, and `facets` break the set down by status, priority, agency, theme, and GAO topic. Counts from free-text queries are approximate (semantic matching); pure filter or agency lookups are exact. Recommendations have no per-document citations, so it is very important to always include the `govquery_url` for the full set in Rec Spotlight rather than citing individual rows.

pia_oversight_recommendations