Dicaria structures primary public disclosures into facts an AI agent can cite: every conclusion carries its reasoning, a verbatim quote and a source link — the source opens, the mirror re-checks, the conclusion survives audit.
Primary public sources across 10+ countries and regions — company disclosures, qualifications, healthcare, legal & sanctions. The full source registry ships with your free allowance.
Ask "does this supplier hold a valid qualification" — generic AI answers fluently; we answer auditably. The difference is not style, it is evidence.
Dicaria// Ask both the same question: "Does this supplier hold a valid qualification?"Fluent answer, no receipt
It can produce a plausible answer — but the quote may not exist, and generated text cannot be checked verbatim against the original disclosure. Where coverage ends, it stays silent and leaves you guessing.
Field structure matches production deliveries; production records (real entities, quotes, hash values) ship with the free allowance — sign up and they are yours.
Every conclusion carries a direct link to the official gazette — read the original without going through us.
If the original goes offline, our mirrored batches remain — re-check byte-for-byte by hash.
Sign up and get 2,000 reads — run your own questions and lists and see the output shape for yourself.
When nothing is found you get a coverage statement: what the pool holds, what it does not — not covered ≠ no risk.
One qualification pool currently holds 17,874 rows / 8,929 distinct entities across national published batches and regional rosters — this entity is not among them.
When nothing is found you get a coverage statement: what the pool holds, what it does not — not covered ≠ no risk.
Full coverage_assertion fields ship with the free allowance.
Four tasks for four kinds of buyer — each card states whether that tier is live or still in preview.
Before onboarding, check qualification identity and validity — certified or not, still valid or not — every row a conclusion with a source link, historical batches preserved.
Procurement-intent discovery: buyer-published intent announcements, filterable by buyer and category — experimental seeds, not full coverage.
Customer / supplier / subsidiary edges carry the verbatim disclosure quote, the disclosure period and a direct announcement link, both directions — no inferred supply chains.
CLI detail tables straight out (table/csv/jsonl), a stable four-part structure agents can parse — entity, reason, evidence and source on every row.
We do not compete on database size: coverage is not provability — every source runs the same pipeline, every file enters the ledger.
Raw files land by domain/batch and every file is logged (URL, sha256, fetch time) — a batch without a full log never enters the pipeline.
Parsed into controlled-vocabulary rows and released only through the QA gate — spot checks, field discipline, polarity tagging; blocked fields never go live.
Auto-merge only on matching legal identifiers; name matches become review candidates — two entities coexisting beats one wrong merge.
CLI detail-table export: qual (conclusion + source columns, batch history re-checkable) / chain (customer / supplier / subsidiary, verbatim-quote column included).
RxNav / Europe PMC / GUDID / ClinicalTrials.gov / DailyMed — the SDK calls official public APIs from your machine, keyless.
Every row: reasoning + verbatim quote + source link; mirror sha256 re-checkable — a takedown does not affect local verification.
One command to install; the free allowance comes with sign-up.
One command into the environment your agents already use: Python 3.10+, pure standard library, zero dependencies.
pip install https://github.com/dicaria/agentdata/archive/refs/tags/v0.3.25.tar.gzCheck version agentdata --version · Update = rerun the command above with the new tag · Uninstall pip uninstall agentdata
Sign up and get 2,000 free reads; search and misses stay free.
Sign up and get 2,000 free reads; search and misses stay free.