Research Agent
Research that returns data, not text you have to parse.
Most tools call a language model with a prompt and hand you back a text blob. The Formt Research Agent searches the web on its own, reads full pages, and returns typed, cited output the next step in your workflow can act on immediately.
How it works
A genuine agent, not a prompt wrapper.
The Research Agent doesn't send one prompt and take the first answer. You give it a brief: the company, topic, or question. It searches the web, reads what it finds, and searches again, following leads until the brief is satisfied or its depth budget runs out. Then it returns the result in the shape your workflow needs: a cited summary for reading, or typed fields the next step uses directly.
At scale
Research every record, not one at a time.
One-off research is what chat AI is for. The reason to put research in a workflow is a list: the agent runs once per record, pulls what's current for each one, and hands typed fields to the steps that act on them.
Output modes
Get back what your workflow actually needs
The output mode decides what the next step can do with the research. Search mode is for human reading. Structured Search is for the steps that come after.
Sources: techcrunch.com · acmecorp.com · linkedin.com
Trust
Autonomous, not unsupervised.
An agent that feeds your CRM and your outreach has to earn it. Every claim carries its source, you decide where the agent is allowed to look, and nothing acts on the results until you say so.
A citation on every field
Structured Search attaches the source to each value it returns. When a number looks off, you check the page it came from, not the whole internet.
You choose the sources
Restrict research to domains you trust, exclude the ones you don't, and cap how old results can be. The agent works inside the fence you set.
Review before anything acts
Put review on the step that uses the research: the CRM write, the send. The run pauses in the Task Manager until you've looked, and bulk approval keeps big batches quick.
Built in
No API keys. No connected services. It just comes with Formt.
Research steps in other tools start by asking for your language-model key and a search service to call. The Formt Research Agent ships with all of it: the models, the web search, the page reading. Add the step and it works.
Nothing to paste
No language-model key of your own, no search-provider account, no token to generate and guard. The agent's whole toolchain is part of the platform, on every plan.
One tool, not a stack
Search, page reading, and synthesis all run inside the step. There's no external service to connect, keep funded, or debug when a workflow stops in the middle of the night.
Researching in your first session
Drop the step on the canvas, write the brief, run. The free plan runs the same agent, with the same depth and filter options, as every other plan.
How it compares
Other tools have LLM steps. This is different.
In most automation tools, the AI step takes a prompt and returns text, and anything more is yours to assemble from parts. Useful, but it isn't research.
| Capability | Other tools | Formt Research Agent |
|---|---|---|
| Setup | Bring your own model API key, and a search service too if the tool lets you wire one in. | Built in. Add the step and write the brief. No keys, no connected accounts. |
| How it researches | A prompt in, one model call, text out. Web access and iteration are yours to build, where they're possible at all. | An agent that searches the web, reads full pages, and keeps going until the brief is satisfied or its depth budget runs out. |
| Research depth | However far one call gets, or an agent you assemble, cap, and govern yourself. | A depth dial per step: Quick, Standard, or Deep (up to 2, 5, or 10 tool calls). Match the thoroughness to the job. |
| Output format | A block of text. Parsing it into usable fields is your problem. | Typed structured fields, or a markdown summary with citations. Usable as-is by the next step. |
| Citations | None. You take the model's word for it. | A source list in Search mode. A citation on every field in Structured Search mode. |
| Domain control | Only what a search service you connect happens to support. | Allow-lists and exclude-lists per step. Keep research on sources you trust. |
| Recency | The model's training data, or whatever a bolt-on search returns. | A recency filter caps how old results can be. Built for news and signals workflows. |
Use cases
What the Research Agent is built for
The Research Agent starts any workflow whose first question is "what do we know about this?" Whatever it finds flows into the steps that act on it.
Account research brief
Research a company before the call: overview, recent news, funding, tech stack, key people. The output feeds AI Generate to produce a structured brief, then lands as a PDF or a HubSpot note.
See templatePersonalised outreach at scale
Enrich a contact list with live company research, then generate an outreach email per contact that references something specific and current, not a mail-merge blank.
See templateCompany deep dive
Pull SEC filings and live financial data, then let the Research Agent add the context the numbers leave out: news, competitors, management moves. AI Generate synthesises it into one report.
See templateMeeting prep brief
Before a meeting, the Research Agent researches the attendee and their company, and AI Generate turns it into a brief: an opener, talking points, things to watch for. In your inbox before the call.
See templateNews brief on a schedule
Monitor a company or topic. The Research Agent pulls recent coverage inside a recency window, AI summarises what matters, and the brief is emailed on the schedule you set.
See templateLead enrichment at scale
Import a list of companies, run a Quick research pass on each one for a short snapshot, and write the enriched list back to Google Sheets. Hundreds of records in one run.
See templateTry it first