> ## Documentation Index > Fetch the complete documentation index at: https://complete-docs-x17s80.allmind.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Document Search — Complete Guide > Find the exact filings, transcripts, research, news, and your own documents — by keyword or by meaning — and hand the results straight to the AI for a cited answer. Document Search is how you find the source material behind your research. From one search box you can look across AllMind's whole library of market documents — US and Canadian filings, earnings call transcripts and slides, press releases, sell-side and ESG research, sector data, news, and investor-relations materials — plus your own uploaded documents. You can search the literal way (find every document that contains an exact word or phrase) or the meaning-based way (find the most relevant passages even when the wording differs), read any result in-app with your terms highlighted, and hand a document — or a whole set of results — to the AI assistant for a cited answer. *Find it in the left sidebar under **Document Search**, alongside Data Room, Grids, and Reports.* ## When to use it Reach for Document Search when: * You need to **find specific source documents** — "the latest 10-K," "every transcript that mentions a plant closure," "recent sell-side notes on this name." * You want to **locate an exact word, name, or phrase** across thousands of documents (a person, a counterparty, a product, a covenant term). * You want a **fast, cited answer drawn only from real documents**, with the sources attached so you can verify every claim. * You're **gathering the raw material** for a deeper piece of work — sources you'll save to a Data Room, analyze in a Grid, or turn into a Report. Use something else when: * You want a quick fact or an open-ended back-and-forth across all your sources → **Chat**. * You have an open-ended question that needs genuine multi-step research and a deliverable → **Agent Studio**. * You want the same questions answered consistently across many companies or documents → **Grids**. * You want a formatted, templated write-up as the main output → **Reports**. * You want to organize and reuse a set of documents as their own analysis universe → **Data Room** (Document Search is how you find documents to put there). ## How to use it Choose **Keyword** to match exact words or phrases, or **AI Search** to search by meaning. AI Search then offers three depths — **Semantic**, **Ask AI**, and **Deep Dive** (see the next section). Keyword is the default. Type a word or phrase (Keyword) or a plain-language query (the AI modes). In Keyword mode you can add several terms as separate "chips" — a document must contain *all* of them to match. Not sure where to start? Pick one of the prebuilt **templates** and it fills in an expert query and the right filters for you. Use the filters to focus on specific **document types** (e.g. just SEC filings, or just transcripts), a **company** (by ticker or name), a **publish-date range**, and type-specific details like year or filing category. Leaving everything unselected searches the whole library (except your private Data Room, which you opt into explicitly). Results come back as a ranked list grouped by document, each card showing the title, source, type, and date with a matching snippet. Meaning-based results show a relevance score; keyword results show how many times your term appears and a corpus-wide total. Select any result to open it. The viewer opens the document with your terms highlighted — PDFs jump to the most relevant page, web pages get a clean reader view, and transcripts read speaker-by-speaker. Step through every match, download the file, or save it. Send a single document or all your results to the AI assistant for a cited answer (the **Ask AI** and **Deep Dive** modes do this automatically). Then continue the conversation, save the best sources to a Data Room, or carry them into a Grid or Report. ## Search modes — exact vs. meaning vs. AI There are four modes, split into a **Keyword** option and an **AI Search** option that expands into three depths: * **Keyword** *(default)* — finds documents that literally contain your word or phrase, with every hit highlighted. Matching is whole-word and case-insensitive, so "report" won't match "reporting." Add several terms and a document must contain all of them. Best for pinpointing exact mentions of a name, term, or phrase. No AI answer; results are ranked by how many times your terms appear, and you also see a library-wide count of total mentions. * **Semantic** — searches by *meaning* rather than exact wording, returning a focused, highly relevant set of passages each with a relevance score. Best for discovering material on a concept or theme when the exact words vary. No AI answer. * **Ask AI** — runs a meaning-based search and then automatically opens the assistant and answers your question using the documents it found, with citations you can click through to the source. Best when you want a quick, grounded answer to a specific question. * **Deep Dive** — runs a broader meaning-based search and produces a fuller, more thorough synthesis across many more documents, with deeper reasoning. Best for a comprehensive, multi-document briefing — it reads more and takes a little longer. Rule of thumb: use **Keyword** when you know the exact term you're hunting for; use **Semantic** to browse by topic; use **Ask AI** for a fast cited answer; step up to **Deep Dive** when the question is broad enough to be worth reading across many documents. Document Search's "Deep Dive" is the deepest *answer setting within Document Search* — it reads across the documents your search returned and answers from those, in a single pass. It is **not** the same as the standalone Deep Research product, and it does not browse the web or run multi-round research. If you need open-ended, multi-step research with a deliverable, use **Agent Studio**. ## What you can search (sources & document types) Document Search covers AllMind's curated library of market documents, plus your own uploads. You can search everything at once or narrow to any combination of these sources: | Source | What it is | Narrow it by | | ----------------------------- | ------------------------------------------------------------------------------------------------ | ------------------------------------------------ | | **SEC Filings** | US company filings (annual reports, quarterly reports, 8-K events, proxy statements, and more) | Company, filing category, year | | **SEDAR Filings** | Canadian regulatory filings | Company, category, year | | **Earnings Call Transcripts** | Quarterly earnings call transcripts | Company, year | | **Earnings Call Slides** | Earnings presentation decks | Company, year | | **Press Releases** | Company press releases | Company, year | | **Investor Relations** | Non-earnings IR material — investor days, conferences, roadshows (transcripts, slides, releases) | Company, sub-type, year | | **Sell-Side Research** | Third-party / broker research, ratings, estimates, and models | Sector, industry, keyword, asset group, category | | **ESG Reports** | Sustainability, impact, and governance research | Company, year | | **Sector Data** | Industry datasets and sector research | Sector, industry, industry group | | **News** | Market and company news | Company | | **Your Data Room** | Your own uploaded documents | Folder | A few things worth knowing: * **Leaving every source unselected searches the whole library** — but your private **Data Room is never included unless you explicitly select it**, so your own documents stay separate by default. * **Your Data Room results are private to you.** Documents you upload to a Data Room become searchable here only after they finish processing, and only for your account. * **Canadian tickers are handled for you** — you don't need to worry about exchange suffixes when searching SEDAR filings or research. * **News by sentiment** (bullish vs. bearish coverage) is available through the News templates rather than a manual sentiment filter. ## Filtering, company scope, and templates * **Company scope** — pick a ticker (by symbol or company name) and Document Search pre-fills that company across all the relevant sources at once, so you don't have to set it per source. * **Date range** — filter by publish date with quick presets (today, this week, this month, this quarter, year to date, past year) or a custom range. It applies to every mode. You can't select a future date. * **Per-source "Only" shortcut** — restrict a search to just one source in a click. * **Reset** — clear all filters; an active-filter count shows how focused your current search is. **Templates** are the fastest way to start. Document Search ships with a large set of prebuilt, expert-written searches, grouped by the kind of work you do — with the most useful surfaced under **Recommended**: * **Research & Analysis** — sell-side research, industry positioning, bull/bear thesis * **Earnings & Guidance** — earnings results, forward guidance, growth drivers * **SEC/SEDAR Filings** — 10-K deep dive, 10-Q analysis, 8-K events, proxy & governance, SEDAR filings * **Investor Relations** — investor days, conference transcripts and slides, capital markets days, analyst days * **ESG & Sustainability** — ESG performance, ESG risk assessment * **News & Sentiment** — recent developments, bullish catalysts, risk flags * **Industry & Sector** — market data, competitive landscape * **By mandate** — Long Only (quality compounders, dividends, capital allocation), L/S Equity (short thesis, earnings quality, relative value, catalysts, insider signals), Fixed Income (credit profile, maturities, liquidity, covenants, recovery), and Private Equity (LBO screening, value creation, add-ons, management diligence) Each template fills in an optimized query *and* pre-selects the right sources and filters, and runs as a Deep Dive. Scope it to a company and it tailors the dates to that company's fiscal calendar. ## Reading and working with a document Select any result and it opens in the viewer on the right: * **Readable in-app:** PDFs, web/HTML pages (in a clean **Reader Mode**, on by default — switch to "Original View" any time), earnings and IR **transcripts** (laid out speaker-by-speaker), and **news** articles. * **Your terms are highlighted** inside the document, PDFs **jump to the most relevant page**, and you can step through every match with next/previous (or the in-document find box on PDFs). * **Download** the open document — PDFs download with a light watermark showing your account, transcripts download as text, and news opens the original article. Other file types (such as spreadsheets, Word, or PowerPoint in your own Data Room) can be downloaded even when they aren't rendered in the viewer. * **Save it** — bookmark a single document, or save all current results at once, into your **Data Room** to build a reusable set of sources. * **Copy as citation** — in Keyword mode, copy a result's snippet as a clean, formatted citation (quote, source, date, title). ## Bringing in the AI Document Search is a natural on-ramp to the AI assistant, in three ways: 1. **Ask AI / Deep Dive modes** automatically open the assistant and answer your query using the documents the search found — grounded only in those documents, with clickable source citations. 2. **"Analyze with AI"** sends *all* of your current results to the assistant at once and asks your search question for you (available in the meaning-based modes). 3. **"Add to AI Chat"** on an open document loads that single document's full text into the assistant for focused follow-up questions. In all three cases the assistant answers **strictly from the documents you gave it** — it doesn't go off to the web or run a broader research job — so the answer stays scoped to exactly what you searched, and every claim traces back to a source you can open. Comparison questions come back as tables with figures tied to their source rows and dates. ## Tips & best practices * **Match the mode to the task.** Hunting for an exact term? Keyword. Exploring a topic? Semantic. Want an answer? Ask AI, or Deep Dive for breadth. * **Keyword is literal.** It matches whole words, so search the root term you expect to appear, and remember that multiple terms must *all* be present in the same document. * **Narrow before you go deep.** Setting a company, a couple of sources, and a date range makes both the results and the AI's answer sharper — especially for Deep Dive. * **Start from a template** when you're not sure how to phrase a research question; then tweak the filled-in query. * **Save as you go.** Bookmark the documents worth keeping into a Data Room so your sources are ready for the next step (a Grid, a Report, or further analysis). * **Verify with the citations.** The AI answers are grounded in real documents — click the sources to confirm the numbers before you rely on them. ## Limits & things to know * **Keyword matching is whole-word and case-insensitive** — "report" will not match "reporting," and multiple terms must all appear in the same document. This is the usual reason a keyword search returns fewer hits than expected. * **Meaning-based searches return a focused set,** not everything — Semantic returns the most relevant passages, Ask AI reads up to about 30 documents in full, and Deep Dive reads across roughly 100+ documents (the most relevant in full, the rest summarized). If a search returns far more than the AI can read, the lowest-ranked documents are set aside, and the answer notes that some were omitted. * **The AI modes are single-pass and document-grounded.** They answer from the results your search returned — they don't run multiple rounds or browse beyond those documents. * **Relevance scores appear only in the meaning-based modes;** keyword results show match counts instead. * **Dates filter by publication date,** and you can't pick a future date. * **Reader Mode is for web pages,** not PDFs, and depends on the page being cleanly parseable; otherwise you'll see the original view. * **PDF downloads are watermarked** with your account details. * **Your recent keyword searches** (the last 20) are remembered in your browser for quick re-runs. * **Your Data Room is opt-in** and private to you; it's excluded from a "search everything" run unless you select it. ## How it works with other features Document Search is the sourcing layer of the research workflow, and it connects to the rest of the platform on both sides: * **Data Room** — search your own uploaded documents as a source, and save search results into a Data Room to assemble a reusable set of sources. Documents you add to a Data Room become searchable here once processed. * **Chat** — the Ask AI and Deep Dive modes *are* an entry point into the assistant, and "Analyze with AI" / "Add to AI Chat" hand documents to Chat as cited context. It's the same assistant you use everywhere. * **Agent Studio** — an agent can be given the ability to search this same document library during its runs, so it can find and cite source documents on its own. * **Grids** — grid cells can be filled from this same document library, so a document you find here can resurface as a sourced answer in a Grid. * **Reports** — report-building draws on the same library to gather cited evidence, so your sources flow through to the finished write-up. In short: **Document Search finds and verifies the sources; Data Room organizes them; Chat, Agent Studio, Grids, and Reports turn them into analysis and deliverables.** ## Example workflow: from sources to a finished deck A real multi-feature journey — "analyze the gold sector and build a presentation": 1. **Find the sources** — in **Document Search**, use the templates and filters to pull the right material on each sub-topic (supply/demand, major producers, cost curves, macro drivers, price outlook, risks): sell-side research, sector data, filings, transcripts, and recent news. 2. **Verify and save** — open the key documents to confirm they're on point, then **save the best results into a Data Room** so the whole source set lives in one place. 3. **Organize** — keep the project's documents together in that **Data Room** as its own analysis universe. 4. **Research in depth** — fire off **Agent Studio** runs (or Deep Dive answers) on the meatier sub-topics, grounded in those sources. 5. **Structure it** — build a **Grid** to answer the same key questions consistently across the producers or topics. 6. **Synthesize** — generate a **Report** from a template, or have an agent produce the deck — every figure traceable back to a document you found in step 1. The result: a sourced, structured analysis and a finished presentation, assembled across the platform — and grounded in real documents from the very first step. ## Common questions Keyword finds exact words or phrases (no AI). Semantic finds by meaning and ranks results (no AI). Ask AI runs a meaning-based search and writes a cited answer from the results. Deep Dive does the same but searches more broadly and synthesizes across many more documents. Keyword matching is whole-word and case-insensitive, so "report" won't match "reporting," and if you entered several terms a document must contain all of them. Try fewer or broader terms, or switch to Semantic to search by meaning. No. Deep Dive is Document Search's deepest answer setting — it reads across the documents your search returned and answers from those in one pass. For open-ended, multi-step research with a deliverable, use Agent Studio. AllMind's library of US (SEC) and Canadian (SEDAR) filings, earnings call transcripts and slides, press releases, investor-relations material, sell-side and ESG research, sector data, and news — plus your own uploaded Data Room documents when you select them. Yes — use Ask AI or Deep Dive, or click "Analyze with AI" on a result set. The assistant answers from the documents it found, with citations you can open. They're grounded only in the documents your search returned, and every claim carries a citation back to its source — so you can verify each number directly rather than taking the answer at face value. Yes. Select "Your Data Room" as a source to search your uploaded files. They're private to your account and are only included when you explicitly choose them. Bookmark a single result, or save all current results at once, into a Data Room. That builds a reusable source set you can analyze further or turn into a Grid or Report. Yes — PDFs, web pages (in a clean reader view), transcripts, and news open right in the viewer with your terms highlighted and the most relevant page in focus. Start from a template. Pick one that matches your goal (earnings, filings, research, ESG, by mandate, and more) and it fills in an expert query plus the right filters; scope it to a company and it tailors the dates too. ## Getting help For help using Document Search or to talk through a research workflow, reach AllMind support through the in-app support option or your account team.