QuantyDocs

Research companies with an agent

You have a sheet of 40 target companies and need each one enriched: what they do, headcount signals, anything newsworthy. Instead of prompting one-off, you build a reusable research agent and let it work through the rows as budgeted tasks — each run audited, priced and stoppable.

What you need: a sheet with a company-name column, and an active subscription (agent runs spend from your organization's AI budget).

Step 1 — Build the agent

Open the Agents page and create a new agent. The fast path: describe what you want in plain language and let Quanty draft it — the generator proposes a name, description, instructions and starter prompts, which you can edit before saving.

An agent has:

  • Instructions — the persona and method, e.g. how to research and how to phrase findings.
  • Model — which LLM it runs on.
  • Connectors — optional integrations the agent may use (Gmail, Google Sheets, Outlook, Slack, Notion, HubSpot).
  • Sample prompts — up to 6 starter prompts shown in chat; you can also generate these with one click.

Agents are on by default; the is active switch disables one without deleting it.

Step 2 — Point it at your sheet

Open the sheet, then open chat. In the chat header, switch from the default Quanty AI Agent to your research agent. The context banner confirms it's working with your sheet and can add rows, columns and edit cells from chat.

Step 3 — Kick off the research

Ask for the enrichment in plain language:

For each company in this sheet, research what they do and fill the Notes column.

Behind the scenes each unit of work runs as an agent task with a per-run budget — $0.50 by default, capped at $2.00. The task returns a value plus the agent's confidence and reasoning, and the cost of the run.

Step 4 — Watch the Agent activity panel

The live activity panel (in the grid toolbar and the chat header) shows every run with its status: Queued, Running, Done, Failed, Stopped or Budget reached. Values verified by execution carry a "Computed in the Python sandbox" badge; estimates are marked "Not sandbox-verified — model-estimated value".

Anything running can be stopped before its next billable call.

Step 5 — Inspect runs and costs

Open any run to see its full step trail — what the agent looked at and why it concluded what it did. Run costs are recorded against your organization's AI usage, so they appear in the token ledger under Settings → Billing alongside chat and extraction spend.

Step 6 — Review the results

Enriched values land in the sheet like any other AI output: open the cell drawer to see reasoning, approve or reject, and lock anything that must not change as ground truth.

Where to go next