> ## Documentation Index > Fetch the complete documentation index at: https://docs.fixaeo.com/llms.txt > Use this file to discover all available pages before exploring further. # Agents > Build AI agents that do marketing work over your own FixAEO data, and review every result before anything is used. Agents let you build a repeatable job that reads your own FixAEO data, does a piece of marketing work, and hands you a draft to review. Where the [AI Marketer](/ai-marketer) chat answers one question at a time, an agent runs a fixed set of steps end to end. You'll also see agents called **workflows** in a few places — they're the same thing. They live at the **Agents** entry in the sidebar, under the Action group. The page opens on a tab rail: **Overview**, **All Agents**, **Templates**, **Skills**, **Scheduled**, **Resources**, and **Automation**. Two buttons sit in the header: **New folder** and **New Agent**. **Agents are in beta.** Review AI-generated work before you use it. Nothing is published to your site, sent to a person, or written anywhere external on its own — every result stops for you to approve or reject first. Treat what an agent produces as a draft to check, not an answer to trust unattended. ## Two ways to build **Describe it in plain language.** On the Overview tab there's a **What do you want to build?** box. Type what you want the agent to do, and optionally attach up to three source files to work from — plain text, Markdown, CSV, TSV, JSON, HTML, or XML. FixAEO proposes a step-by-step plan, which you can look over and then **Approve & run**. This AI builder is rolling out gradually, so some accounts won't see it yet — if yours doesn't, use the visual builder. **Build it visually.** **New Agent** opens a canvas. You add steps from a side panel, place them, and connect each one to its next step. This builder is always available. The plain-language box has three controls that shape how the agent is built: Chooses which AI model does the work. The default **Auto** uses the best available model; you can pin a specific one instead. The exact models depend on your account. **Auto policy** lets the agent make its own calls as it builds; **Guided policy** asks you before key choices. How hard the model works on a step — **Low**, **Medium**, **High**, or **Max** — plus a **Thinking** toggle. More effort is slower; less is quicker. ## The step library In the visual builder, the side panel groups the steps you can add: * **Research** and **Generate with AI** — investigate a topic, then draft or create the result. * **Capture web page** and **Knowledge base** — pull in a page's content, or draw on approved source material you've set up. * **Your FixAEO data** — **FixAEO citation pages**, **FixAEO brand health**, **FixAEO tracked prompts**, **FixAEO competitors**, and **Past outcome signals**, so the work is grounded in your real numbers. * **Brand kit** and **Audience** — keep the output on-brand and aimed at the right reader. * **Condition** and **Repeat for each item** — branch on a test, or run a step over a list. * **Call an API**, **Send to Slack**, **Create Google Doc**, and **Create WordPress draft** — deliver the result to another tool. * **Human approval** — pause the run for a person to approve, edit, or reject before it continues. You name each step, map its inputs to earlier outputs, and connect the flow. Set the reusable pieces up once on the [Resources](/agents/resources) tab and attach them where you need them. ## Test, save, publish Validates the flow without touching anything external. Writes to Slack, a doc, or an API are skipped, and you get a **Test passed** result with how many steps are ready. Keeps your work in progress. Drafts don't run on their own. Turns the current draft into the version future runs use. Publishing never runs the agent and never contacts anything external — it only marks the version. Every publish is kept in **History**. You can compare the current draft against an earlier version to see what changed, and **Load as draft** to pick one up and edit from there. Runs keep using the version they started with, so editing later never rewrites work that already happened. ## All Agents and folders The **All Agents** tab is your library. Search it, filter by **Location**, **Status**, **Type**, or **Creator**, and favorite the ones you use most. On any agent you can rename, duplicate, archive and restore, or delete it, and organize everything into **folders**. You can also move agents between accounts. **Export** writes a JSON file of the agent's steps only — it leaves out run history, credentials, schedules, and connected accounts. **Import** reads one of those files back in as a new draft, up to 2 MB. ## Run history and what needs attention The **Overview** tab surfaces a **Review or fix N agent runs** panel when something needs you. It lists up to five runs that failed or are waiting on a decision, shown as **Approval is waiting**. Click one to open the run, review the output and the evidence behind it, and act. Every agent also keeps its own past-run history you can search and reopen. ## Sharing Share an agent with a teammate as an **Editor**, who can change, publish, and run it, or a **Viewer**, who can read and follow along only. The owner controls access. See [Team and workspaces](/team-and-workspaces) for how members and roles work. ## Limits and credits Running an agent needs an active paid plan. You can still build, organize, and save without one — when you try to start a run, FixAEO tells you plainly: *"An active paid plan is required to start a workflow. Your saved results remain available."* Across manual, scheduled, and automated runs together, each account can start up to **20 runs per rolling day** and keep **3 runs active at once**. Agent runs draw on the same AI Marketer work credits as the chat, so a heavier task uses more than a light one. See [Plans and limits](/plans-and-limits) and [Usage](/usage) for your allowance and remaining balance. ## Related Ready-made agents and guided recipes you can start from. Reusable instructions you write once and attach to any agent. Run a published agent on a schedule, a webhook, or a bulk list. Reusable building blocks, and the approval flow that gates every run. The conversational sibling that answers one question at a time. Which plans can run agents, and the credit allowance for each.