Platform

Research Engine

Ask a question in plain English and get a reasoned answer with the evidence attached — in seconds rather than an afternoon.

How it works

  1. 1You ask in plain English. No query language, no ticker syntax to memorise.
  2. 2The engine gathers evidence across prices, fundamentals, filings, news and retail sentiment at once.
  3. 3A model synthesises it into a direct answer, rather than a list of links for you to read.
  4. 4Every claim carries its source and a conviction score, so you can check the reasoning instead of trusting it.

Example queries

What is the institutional sentiment around ARM Holdings?

Compare AAPL and MSFT on valuation metrics

Which semiconductor stocks have the highest short interest?

Explain the impact of the latest BoE rate decision on the FTSE 100

What it can see

Each answer draws on whichever of these are relevant to the question. Responses cite the specific filing, article or figure behind a claim.

AreaWhat it includesCoverage
Prices & historyDaily and intraday bars, splits and dividendsGlobal equities, ETFs, FX, crypto
FundamentalsStatements, ratios, estimates and revisionsGlobal
FilingsAnnual and interim reports, insider dealingsUS, UK
NewsHeadlines and full articles, deduplicatedGlobal
Retail sentimentDiscussion volume, 24-hour change, rankingUS equities and major crypto
RegulatoryFirm and adviser registers, permissionsUK

Social analysis

Coming soon

Two readings of what a market is saying about an instrument, kept separate because they are not equally trustworthy.

  • Attention — how much a ticker is being discussed across retail forums, and how that has moved in the last 24 hours. Counted, not inferred: a jump from 32 mentions to 314 is arithmetic, and it is available for every instrument.
  • Conversation — what is actually being said, read live and summarised with links to the posts behind it, so a claim can be checked rather than taken on trust.

Attention comes first for a reason beyond cost: it decides whether the deeper read is worth doing at all. A ticker nobody has mentioned has no conversation to summarise, and an eloquent paragraph about silence is worse than no answer.

Both arrive with the same deterministic checks the rest of the platform uses — thin volume, missing citations, and a single account dominating the discussion, which is the cheapest signature a coordinated push leaves. Social readings are context and never a trading signal: they cannot reach a simulation, because a backtest that reads a live feed stops being reproducible.

Using the CLI

alphonce research query "Is NVDA overvalued?" --format json
NameDefaultDescription
--format"markdown"Output format: markdown, json, or plain.
--sourcesRestrict the evidence used: news, social, fundamentals, technical.
--saveSave the response to a file path.

Auditable by design

Every AI decision is logged with the evidence it used. Reasoning traces export as JSON, so a compliance review can reconstruct why an answer was given rather than take it on trust.