Skip to content

Financial-grade retrieval for AI agents.

Real-time financial news and cited research — one API, one key, built on our own retrieval engine.

Start freeTalk to us

Sign up and you have an API key with $20 of credit — no card, no monthly fee. Top up only when you use it up.

Two surfaces, one API.

One key, one credit balance, one MCP endpoint. Pick the surface that answers the question you actually have.

  • Search

    Real-time financial news.

    Whole articles, not chunks. Point-in-time correct — ask what the corpus knew on any past date.

    $3 / 1k requests

    prepaid credits, same price for everyone

    Read the docs

  • Research

    Answers with citations.

    An agent that reads the filings and the news, then shows you every source it used. Streams as it works.

    up to $5 / turn

    billed by tokens used · reading a finished turn is free

    Read the docs

Three lines to your first citation.

A real query against the news corpus, returning the source link, the matched passage, and the exact credit cost of the call.

curl -sS https://api.ariselabs-search-api.com/v1/search \
  -H "Authorization: Bearer $ARISELABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "TSMC advanced packaging CoWoS capacity 2027",
    "limit": 3,
    "indexed_after": "2026-08-01T00:00:00Z",
    "snippet_chars": 240
  }'

Response (excerpt)

{
  "object": "search_result_page",
  "livemode": true,
  "data": [{
    "object": "news_result",
    "id": "doc_AgFLd3ZQb0h5TmxSc1ZqWnhLd0RtQ2dFdVN0WWJQaU5rMno3RjRhOHhSMWNKdVdlNVE",
    "source": "news",
    "relevance": 1.0,
    "title": "TSMC to more than double CoWoS capacity again in 2027, suppliers say",
    "title_source": "publisher",
    "url": "https://www.digitimes.com/news/a20260805pd211.html",
    "domain": "digitimes.com",
    "publisher": "DIGITIMES",
    "language": "en",
    "published_at": "2026-08-05T02:14:00Z",
    "indexed_at": "2026-08-05T04:51:37Z",
    "text_access": "full",
    "snippet": {
      "text": "…TSMC plans to expand CoWoS monthly capacity to roughly 130,000 wafers by the end of 2027, up from about 75,000 exiting 2026…",
      "matched": true
    }
  }],
  "has_more": true,
  "search_metadata": {
    "query": "TSMC advanced packaging CoWoS capacity 2027",
    "source": "news",
    "as_of": "2026-08-07T13:45:22Z",
    "as_of_mode": "current",
    "indexed_after_effective": "2026-07-31T13:45:22Z",
    "returned_count": 3,
    "degraded": [],
    "exhausted_reason": null,
    "elapsed_ms": 148
  },
  "usage": { "credits": 4 }
}

Requires a card — the free Playground runs the same query in your browser. Try it →

Already in your editor.

A remote MCP endpoint with OAuth, plus local stdio packages for networks that will not let an agent dial out. Six tools, the same key, the same credit balance.

  • one command

    Claude Code

    claude mcp add --transport http ariselabs https://mcp.ariselab.ai/mcp

  • one file

    Cursor

    ~/.cursor/mcp.json → { "ariselabs": { "url": "https://mcp.ariselab.ai/mcp" } }

  • one command

    Codex

    codex mcp add ariselabs --transport http https://mcp.ariselab.ai/mcp

  • any client

    Remote MCP

    https://mcp.ariselab.ai/mcp · registry: ai.ariselab/arise

MCP and the local packages need a card — same $0 a month. How authorisation works.

Three numbers we can prove.

No head-to-head accuracy chart in v1: we do not have a defensible evaluation set yet. These are ours to verify instead.

11

regulators' filings read by Research, cited with regulator-portal links

p95 < 500 ms

end-to-end, measured weekly — public latency graph

—

point-in-time history depth

TODO(content)

首页不许印一个我们兑现不了的历史起点年份。

这一格要等新闻语料的版本历史起点配置真正落地、并实测出「最早可用的 as_of」之后才能填;在那之前印任何年份,远古 as_of 都会返回一个 权威的空结果 —— 那是最坏的一种错:看起来像答案。

We wrote the retrieval engine.

Most search APIs are a thin layer over someone else’s vector database. We built ours: hybrid dense + lexical retrieval, object-storage-native, with the index living on ordinary object storage instead of a per-gigabyte vendor bill.

That is not a vanity project. It is why Search costs $3 per thousand requests rather than the seat-licence economics of the incumbents — there is no per-query rent in our cost structure, and our storage bill follows object storage’s curve, not a vector database’s.

We describe the shape of our cost structure, not a per-query multiple against anyone else’s price list.

Built for people who get audited.

  • Point-in-time by design

    Version-level as_of: you get the version of each document that was valid at that instant, not everything published before that date. No look-ahead bias in your backtest.

  • Every answer cites

    A source URL on every row. No synthesised facts without a document behind them, and the primary call to action on a result is “open the original”.

  • SOC 2–Aligned Controls

    Independent audit in progress. We do not display a badge we have not earned — read the security overview and judge the controls yourself.

EU DPA & SCCs · No training on your queries · Zero data retention on request

Trust Center

Questions we actually get.

How is this different from a general web search API?
They index the open web. We index financial primary sources: a financial news corpus, plus 11 regulators’ disclosure systems that Research reads, with point-in-time versioning. If your question is “what did this company file, and when did we know it,” a general web search API cannot answer it.
Do you return full article text?
Search returns the title, a short extract, and a link to the publisher; the documents endpoint returns the full text, at no credit cost. Every result keeps its link to the primary source — we are a way to find it, not a way to skip it.
What does as_of actually mean?
It selects the version of each document that was valid at that instant, not documents published before that date. An article revised in 2026 will return its 2024 text if you ask as_of=2024-06-30.
Is it fast enough for an interactive agent?
Median search latency is under 100 ms server-side. Research runs as a server-side session: it streams events as it works, keeps running if you disconnect, and completes inside a hard wall-clock budget. Live latency.
Do I need a card to get an API key?
No. Signing up gives you an API key and $20 of credit — no card, no monthly fee, nothing expires. Every request is priced in credits and drawn from that prepaid balance; rate limits and prices are the same for everyone. When the balance runs out, requests stop with a 402 — nothing silently bills. Topping up is a one-time payment (from $5); saving a card just makes the next top-up one click, and lets you turn on optional auto-reload. A card never changes your limits, and there are no hidden charges.
Can I use this in production?
Yes. Self-serve keys carry no formal SLA today — we would rather say that here than in a procurement call. Enterprise contracts carry a 99.9% monthly uptime commitment with a service credit schedule. Every key gets per-key rate limits visible in response headers and a public status page.
Do you run your own infrastructure?
Yes. The retrieval engine, the index, and the crawlers are ours. No third-party search vendor sits between you and the corpus.
Where does my data live?
Queries are served from our US region. We do not train on your queries, we sign an EU DPA with SCCs, and we can turn off retention entirely on request. We do not offer regional data residency today, and we would rather tell you that here than in a procurement call.