Superfunds

Get comprehensive performance and risk metrics for a specific fund. Includes total returns (MTD through 5Y), annualized returns (CAGR), and risk metrics: volatility, Sharpe ratio, Sortino ratio, VaR, CVaR, and max drawdown with dates. Returns: Full metrics profile: returns, annualized returns, and risk data. market="us" adds a `flows` block — creations minus redemptions over 1M/3M/6M/1Y in dollars and as a percent of AUM, which no Israeli fund reports. Its MTD and 6M returns are null: the feed publishes neither.

get_fund_metrics

Show one fund's full profile as an interactive card — use whenever the user wants to see or inspect a single fund. Details + 1Y return & Sharpe.

get_fund_profile_card

Show a specific curated category/theme to the user as an interactive card (theme stats + top funds). Use this whenever the user wants to see or explore a named category, e.g. "show me the S&P 500 category" (resolve the slug via list_curated_categories first if needed). Returns the category metadata plus the full fund membership list with per-fund detail (name, manager, AUM, fees, classification). Use after list_curated_categories to inspect a specific theme. Returns: Dict with the category card and a funds[] list for membership. Returns None if the slug doesn't exist or the card is unpublished.

get_curated_category

All documents from a single source within the window — for single-voice deep-dives (e.g., all of Modi Shafrir's posts to verify a framework before quoting).

get_documents_by_source

Filter values that exist in one market, with the count behind each. Call this once before filtering, so theme_tags / strategy_tags / geography_tags are passed values that exist rather than guesses. Every value returned has at least one row behind it — a tag the classifier is allowed to write but nothing carries is omitted, because offering it would send you to a permanently empty answer. The four tag dimensions share ONE vocabulary across both markets — the same classifier and the same canonical keys — so "defense" means defense everywhere. On the Israeli side they label funds and filter every fund tool. On the US side they label two populations, returned as two groups: registrations ("filings", for list_upcoming_fund_launches) and trading funds ("universe" theme / strategy, for query_funds(theme_tag=..., strategy_tag=...) and as "theme:<key>" / "strategy:<key>" categories). fund_count is funds for market="il"; for market="us" it is filings in the "filings" group and trading funds in the "universe" group. The Hebrew label is the display string for Israeli surfaces; the key is what the tools take. Returns: market="il" — a flat dict of the four tag dimensions, each a list of {en, he, fund_count} ordered by count (all funds-scoped): { "theme": [...], "strategy": [...], "geography": [...], "geo_theme": [...] } market="us" — TWO separate groups, because US holds two non-comparable populations. NEVER compare a count across the groups: { "filings": { theme, strategy, geography, geo_theme }, "universe": { asset_class, category, focus, niche, theme, strategy, asset_class_major, fund_type, index_provider } } "filings" counts in-registration filings (filter with list_upcoming_fund_launches); "universe" counts live trading funds (filter with query_funds).

get_filter_vocabulary

Fetch detailed information for a specific fund. Includes metadata, classification, exposure profile, and latest AUM. Returns: Full fund profile with fees, classification, exposure, and AUM. For market="us" the `us` block carries the deeper taxonomy, the tracked index and its weighting/rebalancing rules, leverage, the issuer and advisor, and the issuer's fund page; theme_tags and strategy_tags are our own classification, finer than the feed's focus (the same keys query_funds and the "theme:" categories take).

query_fund_details

Get a comprehensive profile for one fund manager. Includes: fund count, total AUM, type split, fee stats, category breakdown with market share per category, and top funds by AUM. Use list_managers first to find the manager_id. Returns: Full manager profile with competitive positioning data.

get_fund_manager_profile

Fetch the daily price/NAV history of one fund. Returns: market="il": trade_date, purchase_price, sell_price, asset_value, rate. market="us": trade_date, nav, nav_adjusted, nav_change, nav_change_pct, premium_discount, shares_outstanding, asset_value (millions of USD), net_flow (millions of USD, US only). The US feed's nav and nav_change_pct are NOT split-adjusted — a 1:10 reverse split reads as a +900% day. Use nav_adjusted for anything that compares two dates, or get_fund_metrics for the published returns.

query_fund_eod

Flat list of every quoted numerical fact (rates, yields, CPI, FX, index levels) extracted from the corpus, with source attribution. Use to ground claims in actual numbers: never assert a rate value or market level without anchoring it to a quoted data point. Returns: Dict with window_days, topic_filter, and a list of data_points each with source attribution and doc_date.

get_data_points

Show the user the latest Israeli-market news as a card feed. Use this for "what's happening / latest market news" questions. Hebrew market insights generated by the Superfunds news agent. Each card covers a market signal (bonds, equities, rates, macro, etc.) with an original Hebrew headline, summary, and linked fund IDs for comparison. Returns: List of cards with topic, headline_he, summary_he, search_query, fund_ids, and timestamps.

get_market_news

Show the user the curated fund categories as a browsable card grid. Use this when the user wants to browse or see categories/themes. Curated collections. These are editorially-curated themes (e.g. "Israeli AI exposure", "long-duration bonds", "manager flagships") with hand-picked or rule-based fund membership. Different from get_market_by_category, which enumerates the raw TASE classification tree. Use this when the user asks about themes / topics / "show me funds about X". Returns: Dict with categories: [{slug, title_he, title_en, subtitle_he, category_group, member_count, fund_type_dominant, asset_class_major, geography, tags, is_leveraged, is_kosher, display_order, refreshed_at, fund_ids}], plus a fund_lookup map for the union of all member funds.

list_curated_categories

List a market's fund managers with their market position. Returns each manager's fund count, total AUM, and market share percentage. Use this to discover managers before diving into a specific manager's profile. Returns: All managers ranked by AUM, each with fund_count, total_aum (millions of the market's own currency), market_share_pct and funds_with_aum. market="us" returns hundreds of brands against Israel's dozens. A fund launched in the last few days can be listed before it is priced, which is what funds_with_aum reports next to fund_count; market share is a share of what is priced. The list is ranked, so the brands that matter are at the top.

list_managers

List the intelligence sources catalog with freshness signals. Use this to orient at the start of a research session — see what sources exist, their cadence, and whether any have gone stale (last_doc_date well past expected cadence may indicate scraper issues or meaningful publishing absence). Returns: List of sources with id, name, source_type, cadence, language, last_doc_date, doc_count_total, and doc_count_last_30d.

list_intelligence_sources

AUM growth scorecard for one fund manager — asset-gathering performance. Use this to answer "is X gaining assets", "who is bleeding AUM", "rank managers by recent asset growth". Returns the manager's current total AUM plus absolute + percentage change across standard windows: 1m/3m/6m/1y/2y. AUM change includes both flows AND market performance — a manager whose book is mostly ETFs will see "growth" partly from market moves on their existing assets, not just new investor money. For pure-flow signal, interpret carefully against the manager's fund-type mix (see get_manager_overview.funds_by_type). Returns: Dict with manager_id, manager_name, aum_now, latest_data_date, fund_count, windows: { "1m": {"abs", "pct", "fund_count"}, ... }, and data_quality_note. market="us" adds net_flow per window.

get_manager_aum_growth

Fee positioning per fund manager — how cheap or expensive each one is. Returns aggregate fee statistics for every manager, optionally narrowed to one category. Use this to find the cheapest provider in a category or compare fee discipline across managers. Returns: List of {manager_id, manager_name, fund_count, avg_fee, min_fee, max_fee} sorted by avg_fee ascending (cheapest first). Fees are percent in both markets — the US expense ratio is converted from the basis points the feed files it in. market="il" also has avg_trustee_fee; a US ETF has no trustee fee, its expense ratio already covering what Israel splits out, so the key is absent there. This is the one market tool with no AUM in it, so the US answer speaks for every trading fund, priced or not.

get_market_fees

Period total return for one fund manager across their fund book. Use this to answer "how did [manager] do recently", "is X outperforming", "rank managers by recent performance". Computed as the simple (unweighted) mean of per-fund returns across all of the manager's funds — each fund counts once, regardless of AUM. NOT money-weighted. Pair with get_manager_aum_growth for size-vs-performance picture. Returns: Dict with manager_id, period_days, fund_count, avg_return_pct (simple mean across funds), min_return_pct, max_return_pct.

get_manager_performance

All categories of one market ranked by total AUM. Use this to enumerate the classification space and see how AUM is distributed across asset categories — useful when you don't yet know which specific category to drill into, and the way to discover the category values the other tools accept. Returns: List of {category, fund_count, total_aum} ranked by AUM descending. market="il": the canonical Hebrew TASE classification main values (e.g. 'מניות לפי ענפים', 'אג״ח חברות שקלי', 'כספית שקלית'). market="us": the level's values, plus level and funds_with_aum — how many of the funds counted carry an AUM. Below asset class each row also carries asset_classes, the parents it spans ("Energy" as a focus is Equity; as a category, Commodities).

get_market_by_category

Top-level snapshot of one fund market. Use this to answer "how big is the market", "how many funds exist", "what's the manager / fund-type composition". A starting point before drilling into managers, categories, or specific funds. Returns: Dict with manager_count, fund_count, total_aum and a funds_by_type mapping. Every AUM figure is in MILLIONS of the market's own currency — shekels for market="il", dollars for market="us" — so the two are the same magnitude and differ only by currency. market="il" types are TASE legal forms (e.g. {"קרן מחקה": 812}); market="us" managers are brands, types are legal structures (e.g. {"Open-Ended Fund": 6090}), and funds_with_aum says how many of the funds counted actually carry an AUM — a fund launched in the last few days can be listed before it is priced, so total_aum covers that many, not fund_count.

get_market_overview

Flexible natural-language search over one fund market. Accepts a single free-text query (Hebrew or English) and returns ranked matching funds with enum attributes. Uses hybrid retrieval (BM25 + semantic + reranker) over English fund cards. This is the tool for a THEME on either market — "uranium miners", "Indian equity", "photonics" — because the theme lives in the card text. A theme that is a vocabulary key ("drones") is an exact filter instead: query_funds(theme_tag=...). market="us" covers every trading ETF; a fund that has closed is never returned. Understood filter dimensions (extracted automatically from the query): - fund_type: mutual / tracking / etf / foreign-etf - manager, asset_class_major, currency_denom - is_leveraged, is_kosher, is_irish - theme: defense, cyber, healthcare, technology, green-energy, financials, banks, digital-assets, bitcoin, etc. - strategy: leveraged, inverse, large-cap, blend, value, growth, target-maturity, bullet, investment-grade, esg, etc. - geography: country code (US/IL/JP), region (Asia/Europe), or theme (emerging/developed/global) For numeric filters (fee, AUM, returns) or sorting, compose with the existing query_fund_eod / get_fund_metrics / compare_funds_by_metric tools on the returned fund_ids. Returns: Ranked funds with full enum attribute payload, per-channel ranking signals, and the parsed search context (showing which filters were extracted and which search_text was used).

search_funds_flex

Filterable document pull with full enrichment (summary, tags, sentiment, key_data_points, markets, asset_types, relevance). Replaces the snippet-only RAG pattern: instead of `search → snippets`, query `give me last 14 days of high-relevance docs from research sources where bonds sentiment is bearish`.

get_intelligence_documents

RANK funds by a performance or risk metric — best/worst returns, most/least volatile, highest/lowest Sharpe. Use this for any "top N" or "best X" question. Do NOT use query_funds for ranking — it filters by attributes only and does not sort by performance. Optionally filter the ranked universe by classification category or fund manager. Returns: Ranked list of funds with the requested metric value.

compare_funds_by_metric

Resolve a category by slug to its current member fund_ids. For SQL-pure categories (category_type='sql'), runs the deterministic filter live and returns fresh fund_ids (always up-to-date with the current fund universe). For dynamic categories (category_type='dynamic'), returns the cached fund_ids from the last build run. Use this to answer 'who is currently in category X?' without needing to list the full /categories endpoint. Returns: {slug, category_type, fund_ids, member_count, source}. source='live' for SQL-pure resolved on-the-spot; source='cached' for dynamic categories from fund_categories table.

resolve_category

Search Israeli market RESEARCH — research reports, analyst notes, central bank publications. Supports English and Hebrew queries. Use this for opinions, commentary, and analysis ABOUT the market — NOT for finding specific funds. To find actual funds, use search_funds_flex or query_funds instead. Uses hybrid retrieval (BM25 + semantic + reranker) over enriched topic-segmented chunks from multiple research sources. Returns: Ranked research chunks with source metadata (doc_date included), relevance scores, and the parsed search context.

search_market_intelligence

Funds with the largest AUM movement (inflows or outflows) over a window. Use this to answer "what's hot", "where is money flowing", "which funds are bleeding assets". By default movement is % AUM change between the start and end of the window, which mixes money arriving with the market moving and favours small funds. For "where is money flowing" on market="us", pass sort_by="net_flow": it ranks by dollars of creations minus redemptions, so IVV taking in $5bn outranks a small fund doubling, and a fund launched inside the window counts (its change_pct is null). Returns: List of trending funds with fund_id, name, manager_name, classification, past_aum (start of window), current_aum (end), and change_pct. market="us" rows also carry net_flow — the creations minus redemptions behind the AUM move, which separates money arriving from the market simply moving — and as_of, the session date of that row's current_aum. Rows need NOT share one as_of: the two-pass late-NAV fill can leave a fund a day behind, so compare as_of before reading the ranking as a single snapshot.

get_market_trending_funds

Top tags by mention count, classified as surging / new heat / fading / steady based on recent vs prior split of the window. Returns: Dict with window_days, compare_recent_days, filter_applied, and a list of {tag, total_count, last_n_days_count, prior_count, trend}.

aggregate_tags

Funds that have been REGISTERED but are not trading yet. THIS IS THE ONLY TOOL THAT SEES A FUND BEFORE IT EXISTS. query_funds and search_funds_flex know funds that trade; this one knows funds that have filed. Reach for it whenever the question is forward-looking — "what new uranium ETFs are coming", "is anyone launching a quantum fund", "who is filing in private credit", "what is Roundhill about to launch" — and whenever a search for a live fund comes back empty on a theme that sounds new, because the answer may be that it has been filed and not listed. Every row starts as a prospectus, NOT a tradeable fund: status "in_registration". Israel's rows are Maya ק131 announcements, the US rows SEC registrations; some never list, some list months later. A US row is removed once its fund trades. An Israeli row stays, with status="trading" and trading_since (first price date) — so market="il" also answers "what launched recently"; use query_fund_details on its fund_id for the rest. Rows are per registered share class, and one filing registers many at once, so ROW COUNT IS NOT A COUNT OF FILINGS OR DECISIONS — e.g. a single registration can land as hundreds of rows. To count issuer decisions, count DISTINCT us.accession (the filing document); for distinct issuers count us.cik. Both ship on every US row under the "us" block. Returns: List of dicts, most recent first, each carrying market and status ("in_registration", or for market="il" "trading" + trading_since). market="il": report_id, fund_id, manager_name, name_he/name_en, fund_type, fees (mgmt/trustee/variable), benchmark (name/ticker), TASE classification (top/main/sub), exposure_profile, currency, description_he, the three tag arrays, prospectus_url, htm_url, filing_date, publish_date, offer window. market="us": series_id, name_en, manager_name (the trust), classification_main/major, is_leveraged, benchmark_name, index_provider, description_en, the three tag arrays, objective_text, prospectus_url, filing_date, last_filed, and a `us` block with cik, form_type and accession. Israeli-only keys (the TASE triple, the fee triple, the offer window) are absent rather than null — a fee that has not been set is not a zero. A US registration is labelled in the weekly run. Until then its classification, description_en and tags are null or empty, so a tag or class filter misses it; name_contains reads the prospectus text itself and finds it the day it is fetched.

list_upcoming_fund_launches

What a US ETF actually holds, and what those holdings add up to. US-listed ETFs only — this tool takes no market argument. Maya publishes an Israeli fund's category and its benchmark, never its book, so there is no Israeli side to this question. Use it to answer "what is in QQQ", "how much of this fund is in semiconductors", "which countries does it touch", "does it actually hold what its name says", and to check a thematic claim against the positions rather than against the marketing. Returns: Dict with fund_id, name, classification_main, as_of, holdings_count, holdings_weight, holdings [{rank, symbol, name, weight}] and breakdowns {sectors, countries, regions, market_cap, economic_development}. Weights are percent. symbol is null for bonds, cash and held funds, so rank is the identity. A short position is a negative weight, and a commodity fund's book sums well under 100 because the rest sits in futures and collateral — neither is an error.

get_fund_exposure