AI Agents

Agents with real jobs, not chat windows.

Volobox agents have handles, skills, personas, memberships, and managers. They answer questions, run playbooks, fill forms, analyze data, join calls, and take on assigned work, under rules you set.

Two kinds of agents

Your assistant, and your team's agents

Your personal assistant

  • Private to you: conversations stay yours until you share one
  • Works with your own context across the workspace
  • Acts with your permissions, and asks you first
  • In your private chat, reaches what the Volobox API offers: each change is a card you approve, and no auto-approval rule covers it
  • Reads your connected Google Drive, OneDrive, and SharePoint in your own chat, read-only
  • Remembers what you tell it, privately, and shows what each answer drew on

Shared team agents

  • Belong to the workspace, with a handle, persona, and skill set
  • Mention them with $name in chat; assign them tasks and workflows
  • A manager approves actions beyond the requester's permissions
  • Join calls by voice, with an avatar, a voice, and other names they answer to
  • Listed in an agent directory: what each may do, and what it did lately

Shared agents are visible to team members, and they never see personal data.

Where they work

One agent, many surfaces

The same agent works 1:1, in team threads, on assigned tasks, inside workflows, beside forms, on support tickets, and in calls.

1:1 chat

Ask questions or delegate actions with streaming answers and approval cards.

Team mentions

Bring an agent into any project or task thread with a $mention. It reads the images you attach, and everyone sees when it picks the request up.

Calls

Add an agent to a Volobox call to answer out loud when named. Once an administrator turns it on, send one into a Google Meet, Zoom, or Teams call to answer or only take notes. It uses its own access and changes nothing from a call.

Learn more

Voice chat

Talk out loud with your assistant or a team agent in your own chat with it. Every line is written in, marked Said aloud, and approval cards still apply.

Task assignment

Assign real work; coding tasks spawn a sandbox coding session.

Workflows

Plain-language playbooks or JavaScript the agent runs step by step, with human decision points.

Form assist

Fill with AI: the assistant drafts complex forms field by field.

Data analysis

Agents run Python in a sandbox and save charts and outputs to your Drive.

Charts in chat

Numbers come back as a chart card with View data, a CSV copy, and click-through to the items. Build an insight like this turns it into one that stays current.

Learn more

Channel agents

Let an agent watch one channel and act on new messages without a $mention: bug reports become tasks, ideas become requests.

Code questions

Ask how your code actually behaves. A read-only agent reads a fresh checkout and answers with the file paths it relied on.

Support tickets

Agents read a ticket’s thread, search past answers, and draft internal notes or replies for the desk to approve.

Learn more

In a chat

Ask, plan, then act

Every agent chat has three modes, so you choose how far the agent goes before anything changes. Sources and artifacts work in all three.

Ask

The agent reads and answers. It makes no change to tasks, meetings, messages, or anything else in the workspace.

Plan

The agent explores, asks what it needs, and writes a plan you review. Approve and run carries it out in Agent mode.

Agent

The agent proposes changes on approval cards, and nothing runs until you approve, unless a standing rule you created covers it.

Chat sources

Keep files, web pages, and Volobox items as a chat’s sources. A Volobox item is never copied: the agent opens it with the asker’s access every time.

Artifacts

The agent writes HTML, Markdown, CSV, or JSON documents and publishes each change as a new version. Save one to Drive or turn it into a Docs page.

Your connected files

In your own chat, your assistant reads your connected Google Drive, OneDrive, and SharePoint, read-only, Word, Excel, and PowerPoint files included.

Skills & commands

Build a role, not a prompt

An agent is a durable role, and skills are the parts it's built from: bundles of tools and instructions like Meeting Coordinator, Project Coach, Docs Curator, Request Handler, Web Reader, Repository Analyst, and Support Desk. Add your own, import and export them as files, and wire parametric slash commands for repeatable prompts.

Builder skills work in your own private chat: Agent Builder sets up agents, Workflow Builder writes and rehearses workflows, Insights Builder builds insights and dashboards, and Workspace Admin handles limits and model settings for the people who administer the workspace. Whether a team agent codes is a setting, not a skill: Writes code and Reviews pull requests.

Knowledge

Your workspace is the context

Docs, tasks, requests, meeting notes, and PRs are automatically indexed. Agents answer with sources from your team's actual knowledge: no manual uploads, no separate RAG project, and the index stays inside your deployment.

Personal memory

Your assistant remembers preferences you share or it notices, each with an Undo. Only you can see it, and you can keep a chat out of it.

Learned notes

When the team settles how a project works, an agent proposes a note, a project admin approves it, and every agent there knows it.

Never past access

A note only reaches people who could open what it came from, and agents refuse to remember sensitive personal subjects.

Bring your first agent onto the team.

We'll configure a shared agent for your real workflows during the demo.