Brazil CVM Filings for AI Agents

You are building an AI agent that answers questions about Brazilian public companies — earnings, risks, governance, quarterly trends. The agent needs Brazil filings for AI agents: structured access to CVM documents, not a browser scraping CVM portals.

apicvm gives agent builders a stable tool surface: resolve a ticker, list DFP/ITR/FRE filings, download originals, and stream page-level markdown through callbacks.

The persona

Typical builders:

Research agents — "What were Petrobras capex trends?" backed by DFP/ITR text
Compliance copilots — trace answers to original CVM PDFs
Cross-border quant tools — add Brazil alongside US/EU data without custom scrapers

Common blocker: CVM filings are PDFs in Portuguese with complex layouts. Raw PDF bytes do not fit LLM context windows, and naive text extraction loses tables.

The problem

Agent loops need three things CVM portals do not provide out of the box:

Ticker-first lookup — PETR4 → company → documents
Typed filters — DFP (annual), ITR (quarterly), FRE (reference form)
Machine-readable text — page-level markdown, not OCR soup

Scraping breaks. Bulk open-data ZIPs lack a per-filing download workflow. Your agent needs an API contract it can call repeatedly.

Solution: apicvm agent flow

User question about PETR4
  → GET /v1/companies/resolve?query=PETR4
  → GET /v1/documents?ticker=PETR4&type=DFP&year=2024
  → Pick document.id from results
  → POST /v1/document-text-extractions (callback_url = your agent server)
  → Receive page markdown callbacks → chunk → embed → answer with citations

Step 1: Resolve company

curl -H "Authorization: Bearer $APICVM_KEY" \
  "$APICVM_URL/v1/companies/resolve?query=PETR4&by=ticker"

Returns company id, name, cnpj, and tickers[] — enough to disambiguate the issuer before listing filings.

Step 2: List filings by type

curl -H "Authorization: Bearer $APICVM_KEY" \
  "$APICVM_URL/v1/documents?ticker=PETR4&type=DFP&year=2024&perPage=20"

Your agent (or orchestrator) should pick an explicit document.id. The API returns 409-style ambiguity errors when filters match multiple companies — it never silently guesses.

Step 3: Extract text for the agent loop

curl -X POST -H "Authorization: Bearer $APICVM_KEY" \
  -H "Content-Type: application/json" \
  -d '{"callback_url":"https://your-agent.example.com/callbacks/apicvm","document":{"id":"<uuid>"}}' \
  "$APICVM_URL/v1/document-text-extractions"

Each callback delivers one page of markdown. Feed pages into your vector store or pass relevant chunks into the LLM context with document_id and page number for citations.

Example agent architecture

┌─────────────┐     resolve/list      ┌──────────┐
│ Agent / LLM │ ────────────────────► │  apicvm  │
└─────────────┘                       └──────────┘
       ▲                                     │
       │         page markdown callbacks     │
       └─────────────────────────────────────┘
              │
              ▼
       Vector store (ticker, type, year, page)

Tool design tip: expose apicvm as three tools — resolve_company, list_documents, enqueue_extraction — rather than one mega-tool. Lets the model plan steps and handle ambiguity errors.

LLM discovery

For agent frameworks that read site metadata, apicvm publishes /llms.txt with endpoint summaries and the recommended flow. Point tool-description loaders at it during agent setup.

Current limitations

Extraction is async via callback — no synchronous "give me all text" endpoint.
Callback URL must be HTTPS in production.
Corpus coverage grows with ingestion — not every B3 company may be available yet.
Original filings are in Portuguese — plan translation in your agent if needed.

Next steps

Read the API docs
Get an API key
Download llms.txt
Extract text from CVM PDFs — callback handler details
PETR4 CVM filings — concrete company example

Ready to integrate?

Get an API key and start querying Brazilian CVM filings programmatically.

Read the API docs Get API key