> ## Documentation Index > Fetch the complete documentation index at: https://docs.heylua.ai/llms.txt > Use this file to discover all available pages before exploring further. # For operators > The vocabulary of a Lua agent in plain terms, what you can change from the dashboard and the desktop app, and what needs an engineer This page is for the person who runs an agent from [admin.heylua.ai](https://admin.heylua.ai) and the desktop app rather than from code: a support lead, an operations manager, an agency's client. After it you can name every part of an agent, change the parts that are yours, tell which changes are live at once and which wait for a release, and brief an engineer. ## The vocabulary | Term | What it is | | ------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | [Agent](/concepts/agents) | The thing end users talk to, with an ID that starts with `agent_`. Each has a hosted chat page at `https://heylua.ai/agent?agentId=`. | | [Persona](/concepts/persona) | The instructions the model reads before every conversation. Each save is a numbered version; one is served. | | [Skill](/concepts/skills-and-tools) | A named group of tools, with instructions on when to use them, written in code by an engineer. | | [Tool](/concepts/skills-and-tools) | One thing the agent can do, such as looking up an order; the model decides when to call it. | | [Knowledge](/concepts/knowledge-and-features) | Documents the agent searches when answering: policies, FAQs, product sheets. Used only while Knowledge Search is on. | | [Feature](/concepts/knowledge-and-features) | A built-in ability you switch on or off per agent, such as Knowledge Search, Web Search, or inquiry forms. | | [Channel](/concepts/channels) | Where end users reach the agent: WhatsApp, Messenger, Instagram, Slack, Teams, email, the website widget, phone numbers, Front, MessageBird. | | [Job](/concepts/jobs) | Code the agent runs on a schedule, with no conversation involved. | | [Workflow](/concepts/workflows) | A multi-step sequence the platform runs step by step; it can pause until a person decides. | | [Run](/build/workflows/operate-runs) | One execution of a workflow, with a status and a record of every step. | | Approval and the Inbox | A workflow step that waits for a person. The desktop app's Runs page lists such runs, each with a card to approve, deny, or edit; its Inbox collects the cards that need you. | | [Agent version](/concepts/releases-and-versions) | A numbered snapshot of what an engineer has put live: primitive versions, persona version, model, default voice. Promoting one makes it live; promoting an older one is the rollback. | | [Sandbox and production](/concepts/environments) | Production is what end users get; the sandbox is where engineers try unreleased code. Both share knowledge, features, channels, and stored end-user data. | | [Space](/concepts/spaces) | A dashboard-only supervisor that routes one conversation across several agents. No versions, no rollback; the model decides the route, so test it. | | [Agent template](/concepts/agent-templates) | A packaged agent from the marketplace, deployed as a new agent or installed onto one you have; its publisher can offer updates you accept or decline. | ## Tasks by surface "Yes" means the surface does the whole job; "—" means it has no control for it. Admin dashboard labels are from the checkout used for this page; anything newer is unconfirmed. | Task | Dashboard | CLI | Desktop app | Needs an engineer | | ------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------ | ----------------------------------------------------------- | -------------------------- | | Create an agent | Yes | Yes | — | No | | [Edit the persona and serve it](/concepts/persona) | Yes: **Publish changes** serves it, **Save as draft** doesn't; **Activate version** re-serves an earlier one | Yes | — | No | | [Upload knowledge documents](/build/add-knowledge) | Yes, the **Knowledge** tab | No; `lua resources` lists, views, and deletes | — | No | | Switch a feature on or off | Yes, the **Getting started** dialog | Yes | — | No | | [Connect a channel](/channels/overview) | Yes, except RCS (REST API only) | WhatsApp, Messenger, Slack, email | — | No | | [Buy a phone number and bind it](/channels/sms-and-phone-numbers) | US messaging registration only; numbers are bought in the CLI | Yes, the `lua channels` menu | — | No | | [Configure the website widget](/channels/web-widget/quickstart) | Yes, the **Chat widget** tab | No | — | For the embed on your site | | Change the model | Yes | Yes | — | No, but see the warning | | Review inquiry-form submissions | Yes, **Customer inquiries** | No | — | No | | Watch conversations and logs | Yes, **Conversations** | Yes, `lua logs` | — | No | | [Approve, deny, or edit a run's approval](/build/workflows/operate-runs) | — | Yes, `lua workflows approve` | Yes | No | | Cancel or retry a run, raise its budget | — | Yes | Yes | No | | [Install an agent template](/marketplace/install-and-apply) | — (**Marketplace** installs skills only) | Yes, `lua marketplace template install`; env values only, no persona variables | Yes, the install form, which also answers persona variables | No | | [Install a marketplace skill](/marketplace/skills) | Yes, **Marketplace** | Yes, `lua marketplace skill install` | — | No | | [Build a Space](/concepts/spaces) | Yes, only here | No | — | No | | Add or change a tool, webhook, job, workflow, or processor | The Builder code editor | Yes | — | Yes | | Set a production environment variable | — | Yes, `lua env production` | — | Yes | | Release or roll back an agent version | — | Yes | — | Yes | Approvals are answered in the desktop app or with `lua workflows approve --approval --decision approve`; the admin dashboard has no approval view. A channel's settings, and disconnecting it, are under **Channels**. ## Changes that are live at once and changes that wait These reach end users on their next message: a persona version you publish or activate, knowledge you upload or delete, feature switches, channel connections, widget settings, and the model; so do an engineer's `lua env production`, MCP server activation, and `lua push` of the persona. None of them wait for an agent version. Code waits. A new or changed tool, skill, webhook, trigger, job, processor, or workflow reaches end users only when an engineer pushes it and then promotes an agent version. A rollback is the same promote pointed at the previous version, and it is immediate; it doesn't undo knowledge uploads, feature switches, channel changes, or environment variables. The model and the persona you set in the admin dashboard are replaced the next time an engineer pushes the agent, because the push sends what the code says. Tell the engineer so they put the same value in the project. ## The brief an engineer needs * The agent ID (from the admin dashboard URL or the agent's page) and the channel. * What the end user sent, what the agent answered, and when. * Whether it was production or a sandbox test. * What you changed yourself (persona, knowledge, a feature) and when. ## Limits * An agent keeps at most 100 agent versions; older ones must be deleted to create more. * Finished runs are kept for 90 days by default; an organization can set 7 to 90. ## Next steps Upload documents and switch on Knowledge Search. Which channels connect from the admin dashboard and how. Approve, cancel, retry, and read a run. Every term, one line each, with the code name.