intradayfeed

Posted on Aug 5

We Open-Sourced 19 A2A Financial Microservices (And Refuse to Fabricate Data)

#a2a #aiagents #fintech #api

The problem: AI agents can't trust most financial APIs

Most financial data APIs were built for humans clicking dashboards. When an autonomous agent calls them, three things go wrong:

Missing data is silently filled with synthetic values — fake EPS, fabricated price targets, placeholder volatility numbers
No A2A protocol support — no agent card, no capabilities endpoint, no task lifecycle
No honesty layer — the API returns 200 OK with made-up data instead of telling the agent "I don't have this"

We built IntradayFeed to fix this. It's a market analysis platform with a native A2A agent API exposing 19 microservices — and when a data feed isn't configured, it says so.

What is IntradayFeed?

IntradayFeed is an A2A-accessible market analysis agent that exposes microservices for:

Live data (10 services): OHLC snapshots, EMA21/VWAP bias signals, flash news sentiment, macro catalyst calendar, conviction ranking, composite GO scores
Predictive tier (9 services): Earnings predictions, price targets, volatility forecasts, sector rotation, insider/institutional flow, options flow, transcript analysis

The predictive services are honestly unavailable until a verified data provider is wired. No synthetic EPS. No fabricated analyst targets. No placeholder strikes.

A2A discovery — start here

If you're a buyer agent (or building one), here's the entry point:

Resource	URL
Agent Card (A2A v1)	`https://intradayfeed.com/.well-known/agent-card.json`
Agent Card (legacy)	`https://intradayfeed.com/.well-known/agent.json`
Capabilities	`GET https://intradayfeed.com/api/agent/capabilities/`
Human + machine docs	`https://intradayfeed.com/agents`
LLM crawler context	`https://intradayfeed.com/llms.txt`

Agent Card

{
  "name": "IntradayFeed Market Analysis Agent",
  "protocolVersion": "0.3.0",
  "url": "https://intradayfeed.com/api/agent/",
  "securitySchemes": {
    "AgentKey": {
      "type": "apiKey",
      "in": "header",
      "name": "X-Agent-Key"
    }
  },
  "skills": [
    { "id": "bias.signal", "description": "Directional EMA21 + VWAP bias for one ticker/timeframe" },
    { "id": "flash.news", "description": "Latest market headline with sentiment and extracted affected tickers" },
    { "id": "catalyst.calendar", "description": "US macro catalyst calendar with impact scores" },
    { "id": "market_snapshot", "description": "OHLC bars + EMA21/VWAP bias snapshot for one ticker" }
  ]
}

The full card lists all 19 skills with input/output modes, tags, and examples.

How to call it

1. Auth

All task endpoints require an X-Agent-Key header:

curl -H "X-Agent-Key: $AGENT_API_KEY" \\
     -H "Content-Type: application/json" \\
     -X POST https://intradayfeed.com/api/agent/tasks/ \\
     -d '{"operation": "bias.signal", "input": {"ticker": "ES", "timeframe": "1H"}}'

2. Bias signal example

{
  "operation": "bias.signal",
  "input": { "ticker": "ES", "timeframe": "1H" }
}

Response:

{
  "task_id": "a1b2c3d4-...",
  "status": "completed",
  "output": {
    "direction": "bullish",
    "confidence": 79.0,
    "ema_21_position": "above",
    "vwap_position": "above",
    "hint": "Price above both 21-EMA and anchored VWAP — constructive structure"
  }
}

3. Flash news with ticker extraction

{
  "operation": "flash.news",
  "input": { "subscribe": false, "tickers": ["AAPL", "TSLA"] }
}

Response includes a real headline, sentiment score, and extracted tickers — no class-action lawsuit spam (we filter those).

4. Honest "unavailable" (no fabrication)

{
  "operation": "earnings.predict",
  "input": { "ticker": "AAPL" }
}

Response:

{
  "status": "completed",
  "output": {
    "available": false,
    "status": "unavailable",
    "error": "No verified earnings-estimates provider configured. Refusing synthetic EPS/revenue."
  }
}

The service completes successfully but tells the agent the truth: the data feed isn't wired yet. The agent can decide what to do — wait, find another provider, or skip.

Security

Auth enforced: X-Agent-Key header required (HTTP 403 without)
Rate limited: ~9 calls per 15-second window per client
Input validated: tickers max 32 alphanumeric chars, watchlists max 40 symbols, payload ~8KB cap
Unknown operations: HTTP 400, never executed
Injection blocked: SQLi, NoSQLi, XSS payloads rejected at the gate

Commerce flow (for buyer agents)

IntradayFeed supports the full A2A purchase lifecycle:

Browse → GET /api/agent/capabilities/ (operations + pricing + payment rails)
Quote → POST /api/agent/quote/ with operations[] and optional negotiation %
Checkout → POST /api/agent/checkout/ (tip % + payment asset/network)
Pay → Open payment_url, then POST /api/agent/payments/confirm/
Run → POST /api/agent/tasks/ for your purchased operation
Tip → POST /api/agent/tips/evaluate/ (performance-based, $100/mo cap)

Payment rails: USDC on Solana (per-request micro-transactions) and NEAR (smart-contract escrow).

Why we refuse synthetic data

Most financial APIs fill gaps with fabricated values because "something is better than nothing." For humans, maybe. For autonomous agents making buy/sell decisions — synthetic data is dangerous.

An agent calling earnings.predict for AAPL expects earnings estimates. If the API fabricates EPS from a stale model, the agent has no way to distinguish real data from synthetic. It will act on it.

Our approach: if the feed isn't wired, say so. The response is available: false with the specific reason. The agent can:

Skip the signal
Find another provider
Flag the gap to its operator

This is the honesty layer A2A needs.

Try it

Agent Card: /.well-known/agent.json
Docs: /agents
Marketplace UI: /agent-marketplace
LLM context: /llms.txt

No signup required to read the agent card. You only need an API key to execute tasks.

IntradayFeed is a market analysis platform — we analyze markets; we do not give financial advice or process trades. Registered office: 128 City Road, London, EC1V 2NX. Terms · Privacy

DEV Community