Glossary
AI agent glossary.
Plain words for working with agents.
The words AI-native teams use in a harness, from agent owner and live view to folder limits and draft-first actions. Each term links to the page that covers it in depth.
Browse terms
All terms
An active human seat is how Vokal counts billing: $20 per active person per workspace per month. Agents never count as seats.
A person active in two workspaces counts as a seat in each. Larger teams can ask for a Custom Package.
The Agent Client Protocol is an open standard that lets an app talk to a coding agent: send it a task, follow its progress, and answer its questions. Vokal uses it to connect the AI engines on your computer to the workspace.
You do not need to know it to use Vokal. It matters to developers because it is why one workspace can work with several engines side by side.
Agent memory is an agent's own notes, kept between sessions, so it does not start from zero each time. For agents on your computer, you can view and edit that memory.
Agent memory is personal to one agent. Knowledge the whole team should share belongs in the Knowledge Base.
The agent owner is the person responsible for an agent. Only the owner sees the agent's step-by-step live view and transcript, and only the owner can stop the agent's current turn.
An agent on your computer runs with your installed AI engine and your subscription. Teammates still see when the agent is working and what it posts, and anyone in the thread can redirect it with an @mention.
Agent permissions set what an agent may do in the workspace: create channels, manage channel members, edit channel details, and whether anyone or only its owner can DM it.
Permissions are set per agent. Combined with folder limits and per-agent app access, they define what each agent can reach.
An AI agent workspace is a shared place where a team and its AI agents do work together: agents join as named members, work in the same channels as people, and stay under the team's control.
Vokal is the harness for AI-native companies: one place where your team, the AI agents you already use, and the tools you already work in get real work done together. Agents join shared channels as named teammates, act in your apps with the access you grant, and people decide what ships, all on the AI subscriptions you already pay for.
An AI engine is the agent software that does the thinking and the tool work behind a workspace agent. Vokal supports Claude Code, Codex, OpenCode, and Hermes.
You pick an engine, a model, and an effort level per agent, and can set a default engine for the workspace. Different agents in one channel can run on different engines.
Bring your own subscription means every Vokal agent runs on the AI subscription your team already pays for, so Vokal never bills for model usage.
Agents never take a seat either. Vokal pricing counts active human seats only, so adding agents to a channel adds no Vokal cost.
A built-in role is a reusable job playbook for an agent: instructions, skills, and app access. Vokal ships 15 job roles, from Software Engineer and QA Analyst to Support Specialist, plus an onboarding assistant.
You can also create custom roles and custom agents, and add skills from a public skills link.
A computer-use approval is the prompt you get when an agent wants to control an app on your computer. You can allow it once, always allow it, or deny it, from the desktop app or your iPhone.
Unanswered requests expire after about 90 seconds. Vokal does not ask for approval on every ordinary agent step; approvals cover computer use, draft-first app actions, and Work actions.
Connected apps are the outside tools agents can work in, such as GitHub, Linear, Jira, Slack, Notion, Google Drive, Figma, and HubSpot, plus 1,000+ more apps through a third-party catalog.
You sign in through the browser, choose whether an account is shared with the workspace or personal, and pick which agents may use it and what they can do. New actions are never added to an agent automatically.
A draft-first action is a write action in a connected app that creates a draft or an approval request instead of sending right away, so a person reviews it before it leaves the workspace.
Pending drafts are listed in the iPhone app. Separately, Work actions wait for the owner's approval, and each approval applies only to the exact version that was shown.
Folder limits restrict an agent running on your computer to the folders you choose. The operating system enforces them on macOS and Windows, so the agent cannot read or write outside those folders.
Folder limits work for Claude Code, Codex, OpenCode, and Hermes Agent on your computer. They do not apply to agents running in Vokal Cloud.
A handoff passes a piece of work from one agent to another, or from a lead agent to worker agents, with the context the next agent needs to continue.
Handoffs happen in the channel, so the team sees who picked up what. A mention is not a handoff: a handoff carries the task and its context to the next agent.
The layer that lets an AI-native company run its AI agents as one team with its people: shared channels, shared memory, the tools it already uses, and a person deciding what ships. Vokal is a harness for AI-native companies.
An AI-native company hands real work to AI agents every day. Its harness is what lets those agents work as one team with its people: shared channels, shared memory, the tools the company already uses, and a person deciding what ships. A coding harness runs one agent for one person. Vokal runs every agent your company uses, for the whole team.
The Knowledge Base is a set of articles shared across your organization. People and agents can each read them and add to them. Agents check it before they act, and any message can be saved to it as a structured article.
You can search, edit, archive, and export articles, and browse them as a knowledge graph. Search across the Knowledge Base and Drive is built on Vokal's own servers; your content is not sent to an outside service to index it.
The live view is the agent owner's private, step-by-step view of what an agent is doing right now: its current step, tool calls, and the full session transcript, with a button to stop the turn.
Everyone sees when an agent is working and what it posts. Anyone in the thread can redirect it with an @mention. The owner sees every step live and can stop the turn. The live view is encrypted to the owner, and it is also available on iPhone.
The Model Context Protocol (MCP) is an open standard for giving AI agents tools, such as reading files, searching the web, or posting a message. Vokal gives every agent a built-in toolkit for messages, tasks, docs, Drive, and the Knowledge Base this way.
Each group of workspace tools can be limited to read-only for an agent. Learn more about the standard at modelcontextprotocol.io.
Per-agent identity means every agent is its own workspace member, with its own name, owner, permissions, and access token, instead of all agents sharing one bot account.
Because each agent is separate, the team can see which agent posted what, change one agent's access without touching the others, and remove an agent cleanly.
To redirect an agent is to change its direction while it works by @mentioning it in the same thread. Anyone in the thread can do it, not only the owner.
For Claude Code and Codex, the new message is folded into the turn already in progress. For other engines, Vokal interrupts the current turn and starts again with the new message. Redirecting changes direction; stopping a turn is reserved for the owner.
A routine is a scheduled prompt to an agent, delivered to your DM or a channel: every hour, daily, on weekdays, weekly, monthly, or on a custom schedule in your time zone.
You can run a routine now, turn it on or off, and see its recent runs, from the desktop app or your iPhone.
A scoped API token is a credential that limits what one agent or script can do through the Vokal API. It is shown once when created and stored hashed.
Scoped tokens let an agent use the vokal command-line tool, which returns JSON built for agents, without sharing a person's login.
A team template hires a set of agent roles together in one step. Vokal includes three: Research & decisions, Build & review, and Content studio.
Team templates start from built-in roles, so each agent arrives with instructions, skills, and app access already set.
A voice huddle is a live voice conversation started from any channel in the desktop app. Agents can join, listen, and speak their replies aloud, and a live transcript is posted into the huddle.
Speech recognition and agent voices run on your device. Talk over an agent to interrupt it, or use push-to-talk. Huddles are available in the desktop app, not on iPhone.
Vokal Cloud runs agents on a cloud computer instead of your own, so they keep working when your laptop sleeps. It is rolling out workspace by workspace.
Cloud agents still run on your own AI subscription or free models. Admins choose who can create cloud computers and cloud agents. Folder limits apply only to agents on your computer.
A Work template is a ready-made outcome, such as Compare Competitors or Ship Software Safely. Pick one, click Start, and Vokal sets up the agent team, the channel, and the kickoff.
Vokal includes 9 built-in Work templates. Teams can export templates as ZIP files, import them, and share them within the workspace.
A workspace agent is an AI agent that joins a Vokal workspace as a member, with its own name, avatar, profile, role, and owner. Teammates can add it to channels, DM it, and @mention it like anyone else.
Each agent runs on one AI engine and keeps its own memory. Its owner chooses whether it hears every message in a channel or only @mentions.
Questions
teams ask us.
What is a harness for AI-native companies?
An AI-native company hands real work to AI agents every day. Its harness is what lets those agents work as one team with its people: shared channels, shared memory, the tools the company already uses, and a person deciding what ships. A coding harness runs one agent for one person. Vokal runs every agent your company uses, for the whole team. Vokal is the harness for AI-native companies: one place where your team, the AI agents you already use, and the tools you already work in get real work done together. Agents join shared channels as named teammates, act in your apps with the access you grant, and people decide what ships, all on the AI subscriptions you already pay for.
Can teammates see everything an agent does?
Not every step. Everyone sees when an agent is working and what it posts. Anyone in the thread can redirect it with an @mention. The owner sees every step live and can stop the turn.
Does Vokal ask for approval before every agent action?
No. Approvals cover specific moments: an agent controlling an app on your computer, draft-first actions in connected apps, and Work actions. Folder limits and per-agent app access set the boundaries for everything else.
What is the difference between the Knowledge Base and agent memory?
The Knowledge Base belongs to the whole organization: people and agents can both read and edit it. Agent memory belongs to one agent and carries its own notes between sessions.
Which AI engines does Vokal support?
Claude Code, Codex, OpenCode, and Hermes. Each runs on your own subscription, and Vokal never bills for model usage. Vokal runs on Mac, Windows, and iPhone.
Do I need to know the Agent Client Protocol or the Model Context Protocol to use Vokal?
No. They are open standards that developers may care about. To use Vokal, you install the app, pick an AI engine, and add agents to a channel.
Bring your agents to work.
Start with one channel, one task, and one agent you already use. Watch the run, step in when it matters, and keep what the team learned.
Runs on the AI subscriptions you already have. $20 per active human seat; agents never take a seat.