What's New
A dated briefing on recent events at a company, with the context an AI assistant needs to explain what changed and why it matters.
Context for AI assistants
A language model knows the world up to its training cut-off, and after that only what it decides to search for. Leeway delivers named, dated context objects instead, covering a stock, a commodity, an industry or the market as a whole.
Connect the MCP NowSee a real answer
A context object is a named, dated object about a security, an industry or a market: current developments, a timeline, competitors, risks or price drivers. Every object sits in a cache shared by all users. An assistant requests it as a tool over MCP or as an ordinary REST call. The first call after the freshness window has passed generates the object; every later call reads it for two cents.
Run the test on your own assistant
Open ChatGPT, Claude or any other model and ask these three questions. Pay less attention to the tone of the answer than to the names and figures it contains.
- Who leads Apple from September 2026, and what buyback programme did the company most recently authorise?
- What happened to tanker traffic through the Strait of Hormuz in spring 2026?
- Where did the thirty-year US Treasury yield stand in August 2026?
Compare the answers with the excerpts below, which reflect the position on 30 June and 1 September 2026. If your assistant hedges, names people who have since moved on, or points to a search that found nothing, it is working in exactly the situation these objects are built for.
Two layers of knowledge, and neither of them is current
The trained layer
Trained knowledge is broad, often remarkably accurate, and runs up to a cut-off date set by whoever built the model. For business models, accounting logic, industry structure and the longer history of a company it is a sound foundation. What happened after that date is simply not in there.
The search layer
Search only partly compensates. It fires when the model suspects that something needs looking up, and it returns whatever ranks highly for the query it happened to phrase. Both are narrow conditions: an event the model has no inkling of produces no query at all, and a report that ranks nowhere stays invisible even when a search does run.
Why both gaps converge in analysis
Ask about the oil price and you get an answer about supply, demand and inventories that sounds complete, but omits the conflict that moved the price this year. The model did not miscalculate. It never learned that there was anything to look up. For an investment decision the difference matters, because the incomplete answer arrives in the same confident register as a sourced one.
What a brief actually contains
The excerpts below come from stored objects and reflect the position on their respective dates. They are evidence of the form of the output, not a current Leeway view of the market.
Apple's pricing power meets structural supply chain vulnerability
Apple faces mounting pressure from storage costs as AI data centers worldwide drive unprecedented price increases. The company has responded by raising prices on Macs, iPads, and other hardware by double-digit percentages in some cases, a measure it avoided even during the COVID crisis. Simultaneously, Apple is lobbying the Trump administration for permission to source DRAM from Chinese manufacturer CXMT, which already sits on the Pentagon's restricted list. The company lacks its own memory manufacturing capacity and remains structurally dependent on external suppliers.
In the second quarter of fiscal 2026, Apple generated $111.2 billion in revenue, up 17% year-over-year. iPhone revenue reached $57 billion, while Services hit a record $31 billion. The company authorized a new $100 billion share repurchase program, which combined with remaining capacity from the previous program brings total buyback capacity to approximately $164 billion.
The planned CEO transition from Tim Cook to John Ternus in September, coupled with repeated delays in the Siri AI rollout, has raised internal and external questions about execution velocity. A data breach at supplier Tata Electronics, in which over 200,000 Apple-related files surfaced on the dark web, has added reputational and regulatory pressure.
A model whose training ended before these events names Tim Cook as chief executive, knows neither the buyback programme nor the cost pressure coming out of the memory market, and still answers a question about Apple in fluent, complete sentences.
- Week of 1 April: Hormuz transit collapses 95%. Following the 28 February killing of Iran's Supreme Leader, Lloyd's-market insurance withdrawals effectively sever 20% of global crude and fertiliser supply.
- 29 April: historic FOMC dissent. The Fed holds at 3.50 to 3.75% with an eight to four split, the widest since 1992. The internal debate shifts from when to cut to whether cuts are appropriate this cycle.
- 22 May: Kevin Warsh sworn in as Fed Chair. Markets pivot to a hawkish neutral stance, pricing in the risk of hikes if core inflation refuses to roll over.
- 18 August: the thirty-year Treasury yield hits a nineteen-year high at 5.33%. The long end breaks higher on supply and term premium rather than Fed policy, pressuring equity multiples globally.
The same gap runs through the macro picture. Anyone unaware of the change at the head of the Federal Reserve, or of the move at the long end, values every equity with the wrong discount rate.
The biggest current events reshaping the Technology sector are an intense global push to regulate and enforce AI, a rapid surge in AI compute and infrastructure deals, major semiconductor supply-chain shifts and government onshoring efforts, heightened antitrust actions aimed at big tech, and a rise in high-impact cybersecurity incidents.
Antitrust, gatekeepers and platform rules: regulators are actively targeting large platforms with conduct requirements and possible gatekeeper designations that affect data use and AI training, especially in Europe and the UK. Competition authorities are also adding conditions that let publishers opt out of content use for AI features.
The sector brief carries the sources it evaluated. On the public route this object takes no language parameter, which is why the excerpt appears here in its original language.
The objects in detail
Each object can be requested on its own. A ticker takes the form of symbol and exchange separated by a full stop, for example AAPL.NASDAQ. The freshness figure gives the default maximum age of a stored piece; if you need something newer, lower that maximum in the call.
Equities
Seven objects covering a single security, each requested by ticker.
Timeline
A five-year chronology of the events that shaped a company, so an AI assistant can explain how its position and investor view evolved.
Competitors
A structured view of the companies competing in the same market, including their roles, relative strengths and weaknesses.
Risks
The material risks facing a company, with context on how they could affect its business and stock.
Price drivers
A structured view of the forces currently moving a stock’s price and the events or expectations behind them.
What's New: Industry
A briefing on recent events and shifts in the industry, providing context for the company’s market position.
What's New: Sector
A briefing on recent events and shifts across the company’s wider sector, providing context for its outlook.
Commodities and currencies
The same set for commodities and currency pairs, requested with tickers such as XAUUSD.COMM. The exact paths are given in the API documentation.
What's New: Asset
A dated briefing on recent events affecting a commodity or currency pair, with context for its current price.
Asset timeline
A dated sequence of recent events and price moves that shows how a commodity or currency pair developed.
Asset price drivers
The forces currently moving a commodity or currency pair, with the supply, demand or market events behind them.
Asset events
Upcoming and recent events that could affect a commodity or currency pair, with their dates and relevance.
Market
Market objects are written by an internal job. The public route only reads them; where no fresh document exists it answers with 404 and generates nothing.
Market today
The most important developments across regions and asset classes for a current view of the market.
Market this week
A briefing on the past week across markets, focused on what changed and why it matters.
Weekly market timeline
A dated sequence of the structural shifts that have shaped the market over recent months.
Connecting your assistant
Both routes use the same access as the rest of the public API.
Over MCP
The public API is exposed as an MCP server, so an assistant can call these objects as tools. Setup for Cursor, Claude and ChatGPT is described step by step: Set up MCP
Over REST
A call is an ordinary GET request with your token in the apitoken parameter.
GET https://api.leeway.tech/api/v1/public/ai/whatsnew/AAPL.NASDAQ?apitoken=YOUR_TOKEN&lang=enA free account gives you a token. The research itself is billed against your wallet. Create an account
A successfully billed response also reports the amount charged in millieuros, whether it came from the cache, the as-of date and a usage reference.
Limits worth knowing in advance
These constraints are part of the product rather than edge cases.
- The three market objects are never generated on the public route. Where no fresh document exists, the answer is 404.
- A company brief requires a quality rating for that company to exist first.
- Industry and sector objects answer with 400 where the company carries no industry or sector classification.
- The language parameter is not available on every route. Price drivers, industry, sector, commodities and the theme timeline do not accept it.
- Every call also counts against your daily allowance of request units. The wallet does not replace that allowance.
What a call costs
Reading a stored object costs two cents. Only the caller who arrives first after the freshness window has passed also carries the cost of the research itself. Failed calls are not billed.
Frequently asked questions
Why does ChatGPT know nothing about the past few months?
Trained knowledge ends at a cut-off date set by whoever built the model. After that the model only knows what a search returns, and that search only runs when the model sees a reason for it. An event it has no inkling of triggers no search at all.
What exactly is a research brief?
A named, dated research object about a security, an industry, a commodity or the market as a whole. It is requested over REST or MCP and sits in a cache that all users read from.
How current is the material?
Every object carries a date, and each type has a default maximum age ranging from one day to ninety days. If you need something fresher, lower that maximum in the call and a new run is triggered.
Do I need a subscription?
Access requires only a token from a free account. The research itself is billed against your wallet, and the Investor and Trader plans include a wallet balance for each billing period.
Does this replace my assistant's own search?
No. Search remains useful for questions nobody anticipated. The briefs cover the recurring questions where completeness and a verifiable date are what matter.
What form does the ticker take?
Symbol and exchange separated by a full stop, for example AAPL.NASDAQ or SAP.XETRA. The theme timeline is requested with a stated thesis instead.
Are the briefs investment advice?
No. They summarise developments, risks and connections, and replace neither your own analysis nor professional advice.
Give your assistant the current picture
Setup for Cursor, Claude and ChatGPT takes a few minutes, and the first call works with a free token.
Set up MCP© Leeway
PWP Leeway UG (haftungsbeschränkt)
PWP Leeway UG (haftungsbeschränkt)
