// live data for agents

the data your agent
can't find on google.

Picabird turns an agent's question into clean, typed records from official sources: clinical trials, recalls, federal rules, energy regulation and SEC filings. Every record links to its source and records when we fetched it.

ask it something ↓ tell us the dataset you can't get

live real public data, fetched when you ask · POST /v1/search

pick an example or ask your own question

This is a demo, so results come from what the sources publish today. They aren't curated examples. It's rate-limited to a few questions a minute.

how it works

The model only translates your question into a query, and it shows you that translation. The records themselves come straight from the source, mapped field by field, with nothing written by a model.

   your agent                     picabird                                   official sources
 ┌──────────────┐   question   ┌──────────────────────────────┐   exact query   ┌───────────────────────┐
 │ "phase 2     │ ───────────▶ │ 1 interpret  (shown to you)  │ ──────────────▶ │ ClinicalTrials.gov    │
 │  oncology,   │              │ 2 fetch live (no stale index)│ ◀────────────── │ openFDA               │
 │  last 30d"   │ ◀─────────── │ 3 normalize  (typed fields)  │    raw JSON     │ Federal Register      │
 └──────────────┘   records    │ 4 cite       (url + time)    │                 │ SEC EDGAR             │
        ▲                      └──────────────┬───────────────┘                 │ FERC (via FR) · more ↓│
        │                                     │ watch (early access)             └───────────────────────┘
        └──────── webhook: record.created · record.updated {diff} ◀──┘

your agent
   │ "phase 2 oncology, last 30d"
   ▼
picabird
   1 interpret  (shown to you)
   2 fetch live from the source
   3 normalize  (typed fields)
   4 cite       (url + time)
   │            ▲
   ▼            │ raw JSON
ClinicalTrials.gov · openFDA
Federal Register · SEC EDGAR
FERC · more ↓
   │
   ▼
records  ·  watch → webhook {diff}

six places agents go blind

The data is public, but it's buried in portals, PDFs and feeds that change daily. Live means you can try it above right now. Early access means we're building it with design partners.

sources, honestly

We only use official APIs and files the publisher makes available, within their terms. We don't get around logins, captchas or bot walls. "Refresh" is how often the source itself publishes.

source	what	refresh (by source)	status
ClinicalTrials.gov API v2	trial registrations, status, endpoints, sponsors	daily	live
openFDA	drug, device and food recalls (enforcement reports)	weekly	live
Federal Register API	proposed and final rules, notices, comment deadlines	each business day	live
SEC EDGAR full-text search	8-K, 10-K and other filings by phrase	as filed	live
FERC via Federal Register	interconnection, transmission and tariff rules and notices	each business day	live
CourtListener (Free Law Project)	federal dockets and filings	continuous	early access
Regulations.gov	rulemaking dockets and public comments	continuous	early access
ISO/RTO interconnection queues, PUC dockets	MISO, ERCOT, PJM, CAISO, SPP and state commissions, pending permission from each publisher	varies	early access
County courts, council packets, customs	state and county filings, agendas, bills of lading	varies	early access

built for the agent reading it

A raw ClinicalTrials.gov record runs to about 34 KB. A Picabird record is a few hundred tokens with the same facts your agent needs.

provenance on every fieldWe give the official URL, when we fetched it, when the source last updated, and the path in the source each field came from.
shows its interpretationWe return how we read the question, for example "last 30 days" becomes updated_since 2026-10-05. An unsupported filter returns an error instead of being silently dropped.
notices for the model, warnings for youNotices are written for the model to read. Warnings are written for the developer.
unknown stays nullWe never infer a value the source didn't publish. Summaries are assembled from fields by code, not written by a model.
watches early accessSave a query and we'll POST new and changed records to your webhook, with a field-level diff, e.g. "primary endpoint changed".
agent-native/llms.txt, JSON or markdown responses (Accept: text/markdown). An MCP server is next.

// one record, trimmed (shape is picabird.record@1)
{
  "id": "ctgov:NCT07539311",
  "kind": "clinical_trial",
  "summary": "Phase 2 · Recruiting · Genmab ·
    advanced GI cancers · primary: ORR",
  "event": { "type": "updated", "date": "2026-11-03" },
  "fields": {
    "phase": ["PHASE2"], "status": "RECRUITING",
    "sponsor": "Genmab",
    "primary_outcomes": ["Confirmed ORR"]
  },
  "matched": [{ "criterion": "oncology",
    "evidence": "Advanced Gastrointestinal Cancers" }],
  "provenance": {
    "source": "ClinicalTrials.gov API v2",
    "url": "https://clinicaltrials.gov/study/NCT07539311",
    "retrieved_at": "2026-11-04T06:40:12Z",
    "source_updated_at": "2026-11-03T10:00:05",
    "paths": { "fields.phase":
      "protocolSection.designModule.phases" }
  }
}

tell us the dataset you can't get

We're early, and we're choosing what to build next with design partners. Tell us what your agent needs and how fresh it has to be. You'll hear back from a person, not a drip campaign.