discover

When the user invokes /discover, follow this process:

1. Resolve names to real handles

Call lookup_brands({ query }). It searches Adology's central brand directory and returns, for each match, the category, website, domain, and whatever handles the directory holds — Instagram, TikTok, YouTube, X, LinkedIn, and the Ad Library page id — already normalized to the shape author_portfolio_context accepts. It is free, reads only, and adds nothing.

A directory entry may carry only one or two channels. That is the real state of the entry, not an error; report which channels exist rather than inventing the missing ones. If hasMore comes back true, the directory holds more matches than the page you asked for.

2. Find who is already doing the thing

When the ask is about content rather than company names — "who is running founder-led video", "which competitors talk about refill packs" — search the data instead of the directory. In a project that already has scope, search_all({ projectId, query }) runs a keyword search over item analysis text and returns the matching posts ranked by engagement, each with its brand and URL. For a meaning-based recall rather than keyword matching, use analyze({ projectId, query, mode: "semantic" }).

The brand names on those results are discovery leads in their own right: someone whose content keeps surfacing is a candidate to track. Feed the promising names back through lookup_brands to get their handles.

Creators work the same way. If a creator is not in the directory, ask the user for the handle and platform — that is enough to track them.

3. Present the candidates

Number the list so the user can answer with numbers. For each candidate give the name, the category, the channels that actually exist for it, and one line on why it matches the ask. Keep content examples in a separate short list, with the brand behind each one named.

4. Add what the user picks

Tracking happens on the portfolio, not on a project. Read the current state first with read_portfolio_context({ portfolioId }), then merge the new entries with author_portfolio_context({ portfolioId, upsertItems }). Each item needs an id, a kind (brand, influencer, search, discussion, trend-term, niche, seo), and its identifiers. A brand also needs a role — own for the user's own brand, competitor, adjacent, or inspiration for the rest — and a handles object keyed by platform (instagram, tiktok, facebook, youtube, adLibrary), or {} when the directory had none. kind is one bare token; role and every other attribute is its own field.