News sources¶

Sources are RSS or Atom feeds. For every lookup, ZeroAI fetches up to 20 enabled sources with at most five requests at a time, merges and de-duplicates the items, and feeds the newest ones to the model. Each response is limited to 1 MB by default.

flowchart LR
    T([Ticker AAPL]) --> S1["Yahoo Finance<br/>...?s=AAPL"]
    T --> S2["Google News<br/>...?q=AAPL+stock"]
    T --> S3["Nasdaq<br/>...?symbol=AAPL"]
    S1 --> M["Merge, drop duplicate headlines,<br/>newest first, keep 15"]
    S2 --> M
    S3 -.->|"timeout or error:<br/>listed in 'errors'"| M
    M --> P[Prompt for the model]

Defaults¶

Seeded on first start; edit or remove freely.

Add your own¶

On the Sources page, or via the API:

curl -X POST localhost:8000/api/v1/sources -H 'content-type: application/json' \
  -d '{"name": "My feed", "url_template": "https://example.com/rss?symbol={ticker}"}'

The three built-in feed hosts are allowed by default. To add a custom host, set ZEROAI_ALLOWED_FEED_HOSTS to a JSON array containing the built-ins and your trusted host, then restart the server. Redirect destinations are checked against the same list, and each response is streamed under the configured size limit. The source count, body size, and request concurrency can be changed with ZEROAI_NEWS_MAX_SOURCES, ZEROAI_NEWS_MAX_FEED_BYTES, and ZEROAI_NEWS_FETCH_CONCURRENCY.

How a feed is fetched and parsed¶

Each source goes through the same pipeline. The point of the explicit error messages is that you always learn why a source failed, not just that it did.

flowchart TD
    A["URL template + ticker<br/>{ticker} replaced, URL-encoded"] --> B["GET with a browser-style<br/>User-Agent and Accept: rss/atom/xml"]
    B -->|"timeout or connect error"| R{"First try?"}
    B -->|"HTTP 5xx"| R
    R -->|yes| B2["Wait 0.3 s, retry once"] --> B
    R -->|no| E1["error: timed out / could not connect / HTTP 5xx"]
    B -->|"HTTP 4xx"| E2["error: HTTP 403, 404 ...<br/>(not retried, it would not help)"]
    B -->|"200"| C{"Looks like HTML?<br/>content-type or body starts<br/>with doctype/html"}
    C -->|yes| E3["error: this is a web page,<br/>not an RSS/Atom feed"]
    C -->|no| D["Parse XML with defusedxml"]
    D -->|"invalid XML"| E4["error: not valid RSS/Atom XML"]
    D -->|ok| F["RSS: every item<br/>Atom: every entry"]
    F --> G["title, link, date, snippet<br/>HTML stripped, 400 chars"]
    G --> H(["NewsItem list"])

Items without a title or an http(s) link are dropped (a feed cannot smuggle a javascript: link into the page); dates without a timezone are treated as UTC.

Why a source can fail¶

Test a feed before you add it¶

On the Sources page, Test feed fetches the URL once with the ticker AAPL and tells you what it found, without saving anything. Every saved source has a test button too.

curl -X POST localhost:8000/api/v1/sources/check -H 'content-type: application/json' \
  -d '{"url_template": "https://www.nasdaq.com/feed/rssoutbound?symbol={ticker}"}'
# {"ok": true, "kind": "rss", "item_count": 15, "sample": ["..."], "error": null}

Behaviour worth knowing¶

• One bad source never fails the briefing. It appears in the errors list (with the reason above) and as a notice in the UI.
• Timeout per attempt: ZEROAI_NEWS_TIMEOUT_SECONDS (default 8). Timeouts, connection errors and HTTP 5xx are retried once.
• Items kept: ZEROAI_NEWS_MAX_ITEMS (default 15).
• Sources per lookup: ZEROAI_NEWS_MAX_SOURCES (default 20); requests in flight: ZEROAI_NEWS_FETCH_CONCURRENCY (default 5).
• Only headlines and feed snippets are used; article pages are not downloaded.