> ## 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.
# Agent Studio — Complete Guide
> Your autonomous AI research analyst — what it does, when and how to use it, the modes, and how it fits into your research workflow.
Agent Studio is your autonomous AI research analyst. You give it a research question or task in
plain language, and it works on its own — searching, reading, and analyzing across markets, filings,
news, your own documents, and the web — then streams back a structured answer with its reasoning and
the sources it used. Unlike a single chat reply, an agent run can carry out genuine multi-step
research and even produce finished deliverables like a slide deck, an Excel model, or a Word
document.
*Find it under **Research Workspace → Agent Studio**.*
## When to use it
Reach for Agent Studio when:
* You have an open-ended question that needs several steps — "analyze the competitive position of
this company," "what are the key risks in the gold sector right now."
* You want a thorough, well-sourced answer rather than a quick fact.
* You want the AI to produce a deliverable (a deck, model, or document) as part of the work.
* You want to kick research off and come back to it — runs keep working on their own.
Use something else when:
* You just need a quick fact or a fast back-and-forth → **Chat**.
* You want to find documents or sources first → **Document Search**.
* You want the same set of questions answered consistently across many documents or companies →
**Grids**.
* You want a formatted, templated report as the main output → **Reports** (an agent run can feed it).
## How to use it
1. Open Agent Studio and choose or create a **workspace** to keep related runs together.
2. Type your research question or task in plain language. Be specific about what you want and any
focus areas.
3. Pick a **mode** for how deep to go (see below).
4. Start the run. The agent begins working and streams its progress, its thinking, the sources it
pulls, and the answer as it builds.
5. You can leave and come back — the run keeps going, and you'll see everything when you return.
6. When it finishes, review the answer and sources. Ask a **follow-up** in the same thread to refine
or go deeper, or download any files it produced.
7. **Cancel** any time to stop or redirect.
## Choosing a mode (how thorough vs. how fast)
Agent Studio offers four modes, trading speed for depth:
* **Instant** — fastest; for a quick answer or lookup.
* **Balanced** — a quick but more capable single pass; a good default for straightforward questions.
* **Agent** *(default)* — the standard workhorse. Handles real multi-step questions, can break work
into parts, and can generate files. Best for most analyses.
* **Deep Research** — the most thorough. It plans the work, runs several lines of research at once,
and digs deepest. Best for complex, high-stakes questions — and it takes the longest.
Rule of thumb: start with **Agent** for real analysis; step up to **Deep Research** when the question
is broad or important enough to justify the wait; drop to **Instant/Balanced** for quick things.
Agent Studio's "Deep Research" *mode* is simply its deepest setting — it is not the same as the
separate Deep Research product. Users mix these up; the assistant should keep them distinct.
## What it can access & produce
**It can draw on:**
* Live market data — prices, fundamentals, estimates, technicals, sector and peer data.
* News and sentiment, events and earnings reactions.
* Filings and research — company filings (US & Canada), earnings-call transcripts, broker research,
ESG, and third-party research.
* Your own connected materials — your Data Room documents, uploaded files, Google Drive,
OneDrive/SharePoint, email, and connected data sources.
* The web.
**It can produce:** a structured written answer with sources, plus finished files — PowerPoint
decks, Excel models, and Word documents — that you can download.
## Tips & best practices
* Be specific about the goal and the output you want ("…and build a 5-slide summary deck").
* For a big topic, split it into several focused runs (one per sub-question) — it's faster and the
results are cleaner (see the workflow below).
* Point it at your own documents (via a Data Room) when you want the analysis grounded in your
materials.
* Use follow-ups to refine rather than starting over — the run keeps its context.
* Match the mode to the stakes; Deep Research is worth the wait for complex work.
## Limits & things to know
* Deeper modes take meaningfully longer — that's expected; they're doing more thorough, multi-step
work, sometimes several lines of research at once.
* The agent answers from what it can access (above) at the time it runs.
* "Deep Research" mode here is not the standalone Deep Research product (common confusion).
* Runs are saved in your workspace so you can return to them; export anything you want to rely on
long-term.
## How it works with other features
Agent Studio is the engine of the research workflow and connects naturally to the rest of the
platform:
* **Document Search** → find and gather the right sources first, then have an agent analyze them.
* **Data Room** → keep a project's documents together and point agent runs at them for grounded
analysis.
* **Grids** → when you want the same questions answered consistently across many documents or
companies, a Grid is the structured complement to an agent's narrative.
* **Reports** → turn an agent's findings into a formatted, templated report as the final deliverable.
* **Chat** → for quick follow-ups and clarification.
It shares the same underlying assistant as Chat, but runs autonomously on a task instead of a live
back-and-forth.
## Example workflow: analyzing a sector and building a presentation
A real multi-feature journey — "analyze the gold sector and build a presentation":
1. **Scope it** — break the goal into the 7–8 key topics (supply/demand, major producers, cost
curves, macro drivers, price outlook, risks).
2. **Gather sources** — use **Document Search** to find relevant filings, research, and news on each
topic.
3. **Organize** — collect the best sources into a **Data Room** for the project.
4. **Research in parallel** — fire off several **Agent Studio** runs, one per topic, to research each
in depth (Deep Research mode for the meatier ones).
5. **Structure the analysis** — build a **Grid** to answer the same key questions consistently across
the topics or companies.
6. **Synthesize** — combine the findings and generate a **Report** from a presentation template, or
have an agent run produce the deck directly.
The result: a sourced, structured analysis and a finished presentation, assembled from across the
platform.
## Common questions
Speed vs. depth: Instant (quick) → Balanced → Agent (standard, multi-step, can make files) →
Deep Research (most thorough, takes longest).
It uses the same underlying assistant as Chat but runs autonomously on a task. Its "Deep
Research" mode is its deepest setting, not the separate Deep Research product.
Deeper modes do more thorough, multi-step work and may run several lines of research at once —
that's normal. Pick a lighter mode for speed.
Market data, news, filings and research, your own documents and connected sources, and the web.
Yes — ask for it, and download the file it produces.
No — it keeps running, and you'll see everything when you come back.
Ask a follow-up in the same run to go deeper or adjust.
Yes — a common pattern for big topics is to launch one run per sub-question (see the workflow
above).
## Getting help
For help using Agent Studio or to talk through a workflow, reach AllMind support through the in-app
support option or your account team.