---
title: "The Agent: an AI agent that asks before it acts | SalexHub"
description: "The Agent answers from your records, drafts emails and changes records with your permissions. By default it asks before every change and every send."
url: "https://salexhub.com/product/agent"
---

# The Agent: Ask for the work. Approve what changes.

The Agent reads, drafts and changes your records through the same tools and permissions you have. By default it asks before every change and every send. How much it may do on its own is your setting.

You ask the Agent which deals have gone quiet. It reads the deals and their activity on its own, proposes a follow-up task for each, and creates nothing until you approve.

[Start free](https://app.salexhub.com/register) · [Open the chat](https://chat.salexhub.com/)

## What you do with it

**Work you can hand over today.** It works in two places: in the app, in a side panel beside every page, and in the chat at chat.salexhub.com, where it takes the whole screen.

- **Answers from your records**: Ask what a quarter brought or which deals have gone quiet. Records, numbers and charts come back drawn by the product, and a chart can be saved as a report.
- **Drafts on the right record**: Replies, follow-ups and reminders are saved as drafts on the record they belong to. A send from your connected mailbox asks first, then waits a moment you can use to cancel.
- **Changes as cards**: Creating and updating records, moving a document to its next step, building properties and workflows. Each change arrives as a card that shows what will change, with Approve and Decline.
- **Autonomy is a setting**: Four modes, from Ask every time, the default, to Full autonomy. The Owner sets how far anyone may go. Changes to access and personal data always ask, and changing someone's role is never the Agent's to do.
- **Pages from the web**: Give the Agent the address of a public page and it reads it, naming the source of every fact it takes from the web. What it reads there is information, never instructions.
- **Your key, your models**: The Agent runs on your organization's own key from Anthropic, Google, OpenAI, DeepSeek or Vercel AI Gateway. The AI Gateway page in Settings shows what each person and each model spends.
- **Domain Skills** (coming soon): Sales and marketing method the platform carries for you: which qualification fits this deal, what is already known, what is missing and what to propose. A Skill applies itself when the facts call for it, and what it finds reaches the record as a proposal you approve.

## Two ways to work

**Two ways to work. One organization.** Work in the app yourself, or tell the Agent in the chat. The records, permissions and approval rules are the same, so you choose per task.

- **The app** (app.salexhub.com): You drive. Pipelines, documents, reports and settings are on screen, and the Agent helps from a side panel on every page. [How the app works](https://salexhub.com/product/app.md)
- **The chat** (chat.salexhub.com): You say what you need. The Agent does the work with your permissions, brings records and approval cards into the conversation, and waits for your approval. [How the chat works](https://salexhub.com/product/chat.md)

One request, two ways. In the app, the overdue invoices fill the screen and the Agent answers in the side panel; in the chat, the conversation fills the screen and an invoice opens beside it. Either way the three reminder emails wait for approval.

Try the chat before you sign up: five messages, no account. [Open the chat](https://chat.salexhub.com/)

## The Agent here

**Every act goes through one registry.** The product's registry of 523 tools is the only way the Agent touches your organization. It finds the tool it needs, reads its contract and runs it as you, so it never sees or changes more than you can.

Tools from the product's registry that the Agent uses here:

- `search.global`: Finds records across the organization by their text. Runs in every autonomy mode.
- `record.list`: Lists the records you can see, with filters. Runs in every autonomy mode.
- `report.aggregate`: Counts and sums records for an answer. Runs in every autonomy mode.
- `email.draft.save`: Saves an email draft on its record. Asks in the default mode, Ask every time; runs from Auto inside the organization.
- `record.update`: Changes a record, as a card you approve. Asks in every mode.
- `email.send`: Sends an email from your connected mailbox. Asks in the default mode; runs from Auto including replies.
- `batch.undo`: Takes back the changes of one answer. Asks unless the mode is Full autonomy.
- `subject.erase`: Erases a person's data from the organization. No agent runs it in any mode; a person does it in the app.

Thirty-eight tools are never the Agent's, in any mode: among them erasing a person's data, changing roles, deciding approvals and creating API keys. People do those in the app.

## Connected

**It works wherever your records are.**

- [Workflows](https://salexhub.com/product/workflows): An AI step inside a workflow runs on your key and spends AI credits from the run's budget.
- [Reports & dashboards](https://salexhub.com/product/reports): A chart the Agent draws can be saved as a report, on the same question.
- [Objects & data](https://salexhub.com/product/objects): It can build objects, properties and pipelines. Below Full autonomy each change waits for your approval.
- [Control & audit](https://salexhub.com/product/control): It never has more access than the person it works for, and approval rules hold its acts like anyone's.

## Control

**Everything it does is on the record.**

- It runs as the person asking: hidden properties stay hidden, and records you cannot see stay out of its answers.
- Each change shows who caused it: you approved it, or it ran without a click.
- Undo takes back an answer's changes and skips records edited since.
- One answer may take up to 12 steps and 5 minutes, within an AI-credit ceiling that depends on the model.
- AI stays off until the organization pays for AI credits or adds its own provider key, and the Owner can switch it off for everyone.

## Recent changes

- 2026-10-02 (The Agent): **Minis use your connected services.** A Mini can read from the MCP servers an admin connected: the tools marked as reads run on their own, and anything else asks first.
- 2026-10-01 (The Agent): **Minis.** A Mini does one job that repeats: it wakes on a schedule, on an event, when a record enters or leaves a dynamic list, or when the pace of a goal changes, and its report arrives in the app or by email.
- 2026-10-01 (The Agent): **What a Mini may do, and ready ones to start from.** A Mini reads and reports, or proposes: every change it wants becomes a card you approve in its thread, and only a Mini held by an agent you hired acts by itself, inside that agent's limits. A ready Mini is made without a model and without AI credits.
- 2026-09-17 (The Agent): **The Agent can read the public web.** The Agent opens a public page at an address you or a search result gave it, and treats what it reads there as information, never as instructions.

All changes: [Changelog](https://salexhub.com/changelog)

## Read more

- [Autonomy and approvals](https://salexhub.com/docs/agent/autonomy): The four modes, and what waits for a person whatever the setting.
- [Connect an AI provider](https://salexhub.com/docs/agent/ai-provider): Add your provider key, choose the models, and see what the Agent spends.
- [Work from the chat](https://salexhub.com/docs/agent/chat): Work with the Agent at chat.salexhub.com, and what stays in the app.
- [Why the Agent asks first](https://salexhub.com/blog/why-the-agent-asks-first): Why asking first is the default, and what you can let it do alone.

---

The same page for people: https://salexhub.com/product/agent. Index of this site for agents: https://salexhub.com/llms.txt