@ar-agents/boletin-oficial

Argentine Boletín Oficial as a structured firehose for AI agents on the Vercel AI SDK 6.

pnpm add @ar-agents/boletin-oficial ai zod

Part of the Arg toolkit — open infrastructure for the Argentine AI agent jurisdiction.

Why this exists

The Boletín Oficial publishes ~1500 normas a day across four secciones — laws, decrees, ARCA/BCRA resoluciones, sociedades, public-sector contracting, judicial notices. Compliance, regtech, legal monitoring, and "did the government just publish something about us" all hit it.

There is no documented public API. There is no RSS. There is no webhook. Every Argentine company that monitors the BO is either scraping the same endpoint, paying a consultancy to do it, or missing things.

This package fixes that for AI agents and developer apps:

• Search + filter by sección, organismo, CUIT, date range, free-text.
• Get a single norma by id with full text + extracted CUITs.
• Subscribe to keywords/CUITs/secciones and get matched normas back.

Pluggable adapters: ship-with-defaults LiveBoFetcher, MockBoFetcher for tests, UnconfiguredBoFetcher for safe-by-default tools, plus an InMemoryBoSubscriptionAdapter and a BoSubscriptionAdapter interface for production stores.

Quick start

import { Experimental_Agent as Agent, stepCountIs } from "ai";
import {
  boletinOficialTools,
  LiveBoFetcher,
  InMemoryBoSubscriptionAdapter,
} from "@ar-agents/boletin-oficial";

const agent = new Agent({
  model: "anthropic/claude-sonnet-4-6",
  tools: boletinOficialTools({
    fetcher: new LiveBoFetcher(),
    subscriptions: new InMemoryBoSubscriptionAdapter(),
  }),
  stopWhen: stepCountIs(8),
});

const { text } = await agent.generate({
  prompt:
    "¿Hubo algo nuevo de ARCA en el Boletín Oficial esta semana? Si hay alguna resolución sobre monotributistas, suscribime para que me avises.",
});

The agent will:

1. Call bo_search with { organismo: "ARCA", from: "<7 days ago>" }.
2. Filter results by relevance.
3. Call bo_subscribe with { owner_id, organismo: "ARCA", keyword: "monotributo" }.
4. Summarize.

Tool surface

| Tool | Purpose | | ------------------------- | ------------------------------------------------------ | | bo_search | Free-text + filters → list of normas. | | bo_get_norma | Fetch a single norma by id. | | bo_today | Convenience wrapper for "today's publications". | | bo_subscribe | Register a keyword/CUIT/sección match. | | bo_list_subscriptions | List active subscriptions for an owner. | | bo_unsubscribe | Remove a subscription by id. |

See AGENTS.md for tool-selection rules.

Direct API

You can also use the building blocks without the agent layer.

import {
  LiveBoFetcher,
  matchNorma,
  classifyTipo,
  extractCuits,
} from "@ar-agents/boletin-oficial";

const fetcher = new LiveBoFetcher();
const { results } = await fetcher.search({ organismo: "ARCA", query: "monotributo" });

// Pure helpers — no I/O, safe in any environment:
classifyTipo("RESOLUCIÓN GENERAL Nº 5612/2026", "primera"); // → "resolucion"
extractCuits("CUIT del responsable: 20-41758101-5"); // → ["20417581015"]

Resilience model

The Boletín Oficial doesn't publish a stable API. Parsers in this package are conservative: when the page structure changes, the package returns empty results or throws BoError("fetcher_unexpected_response") rather than silently return wrong data.

Pin the package version. Watch the changelog. Report parser breakage on GitHub — a fix usually lands within 24h.

Production storage

InMemoryBoSubscriptionAdapter is fine for dev and demos but loses state on restart. Implement BoSubscriptionAdapter against your preferred store:

class PostgresBoSubscriptionAdapter implements BoSubscriptionAdapter {
  async put(sub) { /* INSERT … ON CONFLICT */ }
  async get(id) { /* SELECT */ }
  async list(filter) { /* SELECT WHERE */ }
  async remove(id) { /* DELETE */ }
}

License

MIT © Nazareno Clemente