Skip to main content
A connected assistant works in the workspace you chose when you connected it. You ask in plain words, and the assistant picks the right action. You don’t need to name tools or ids. On a guest account the assistant can do all of this within the guest allowance, except open the dashboard. See Guest access.

Find what you already have

  • List and search your workflows and agents, and open one to see its steps, inputs and settings
  • Show recent runs, with their status, cost and output
  • Show the apps you’ve connected, and give you a link to connect a new one
  • Search the Fetch Hive docs when it’s unsure how something works
  • Tell you what your runs cost over a day, a week, a month or your billing period, by workflow and by agent, and how many tasks and how much hosted model credit you have left
  • Explain what your plan allows, such as steps per workflow or runs at once, when something is refused
  • List the AI models you can use in prompt steps, agents and image steps, and what each one supports
Try asking: “Which of my workflows post to Slack?”

Start from a template

  • Search the template library and your own account templates
  • Explain what a template does and which inputs it needs
  • Copy a template into your workspace, the same as the Use template button
  • Copy a shared workflow or shared agent into your workspace from its share link, the same as Edit this workflow on the share page

Build something new

Describe what you want, and the assistant builds a workflow or an agent and saves it as a draft. If the request is unclear, it asks you a question first and passes your answer back. Building doesn’t use credits. Try asking: “Build me a workflow that summarises new Hacker News posts about AI every morning and sends them to me on Slack.”

Change a workflow

  • Rename a workflow, change its description, or add and remove start inputs
  • Add a step, change a step’s settings or prompt, move a step, or remove a step
  • Apply quick fixes to common small problems in a workflow’s steps. These need no AI and cost nothing
  • Add app actions such as sending a Slack message or creating a Notion page
  • Show the workflow’s saved versions and restore an earlier one
  • Make a copy of a workflow under a new name
  • Save a workflow as one of your account templates
  • Archive a workflow you no longer use, and restore it later
Every change an assistant makes is saved as a new version. See Safety and approvals for how to undo one.

Change an agent

  • Change an agent’s name, description, instructions, model or reasoning effort
  • Add or remove tools, such as web search or an app
  • Let the agent search one of your knowledge bases
  • Make a copy of an agent under a new name
  • Archive an agent you no longer use, and restore it later
  • Make a new webhook secret for an agent or a deployed workflow, for example after the old one leaked. You copy the new secret from the dashboard

Knowledge bases

  • List your knowledge bases and show what’s in one, and whether each source is ready yet
  • Create a knowledge base
  • Add a block of text or a web page as a source, or remove a source
  • Try a search, to check an agent would find the answer
You add files to a knowledge base on the dashboard. See Knowledge Base.

Prompts

  • Search your prompts and open one to see its messages, model, settings and variables
  • Create a prompt, or change its messages, model or settings. Every change is saved as a new version
  • Deploy a prompt so you can call it from your own app
  • Run a deployed prompt with the inputs you give and show you the answer

Run, test and deploy

  • Run a workflow once, as it is now, like the editor’s Run button
  • Send a message to an agent and give you its answer, optionally in a JSON format you choose
  • Follow the run and give you the output of each step
  • Tell you what a run usually costs before you start it
  • Explain why a run failed and what to do about it, run it again, or cancel a workflow run that is still going
  • Give you download links for the files a run made
  • Upload a file, such as a CSV or a PDF, to your Media Library so a run can read it
  • Test a workflow on sample data before you use it for real: steps that read from your apps get sample data, and steps that send messages or wait for a review are switched off
  • Check how a test run went: a score out of 100 and what to improve. If it isn’t good enough, the assistant can fix the workflow and test it again
  • Deploy a workflow so the API, webhooks and schedules use the latest version
  • Give you ready-to-paste code (curl, TypeScript or Python) to call a deployed workflow or an agent from your own app
  • Tell the Fetch Hive team when a run, workflow, agent or build went wrong

Schedule

  • List your schedules and when each one runs next
  • Run a deployed workflow on a schedule, or change when it runs
  • Pause a schedule, and resume it later

Share

  • List your share pages and open one
  • Create a share page for a workflow, with its title, description and input guidance
  • Create a share page for an agent, where visitors chat with it, with its title, description, greeting and starter messages
  • Change a share page’s copy or default inputs, or switch it off
  • Run a workflow share page as a visitor would, to check it works
See Sharing for how workflow share pages work and Sharing an agent for agent pages.

Turn your API into an MCP server

Give the assistant a link to your OpenAPI spec, paste the spec, or let it upload the spec file, and it sets up an MCP Gateway for you:
  • Create the gateway and compile the spec
  • List what needs fixing, such as missing descriptions, unclear tool names or destructive operations, with the fix for each
  • Apply the fixes you agree to, or rename, describe and turn tools on or off itself
  • Set the API base URL when the spec doesn’t have one
  • Publish the gateway, or roll back to an earlier version
  • Try a tool on the live URL and show you the result or the API’s error
  • Read the gateway’s recent calls when something fails
You add your API’s key or token and create MCP client tokens on the dashboard. The assistant gives you the link and never asks for them in the chat. Only account owners and admins can change gateways. Try asking: “Turn https://api.example.com/openapi.json into an MCP server and tell me what needs fixing.” Prefer to run it yourself, or is the API on your own machine? The open-source MCP Gateway CLI is free and does the same from your terminal. The hosted gateway cannot reach localhost, so use the CLI for an API you are still building. Assistants with a terminal, such as Claude Code, Codex or Cursor, can set it up for you.

Pipelines

A pipeline runs several workflows one after another, and stops at checkpoints where a person reviews or chooses before it carries on.
  • Find your pipelines and show their stages, checkpoints and start inputs
  • Tell you what a run usually costs, then start it only after you say yes
  • Follow the run and show you each stage’s result
  • When the run reaches a checkpoint, show you what it asks (a list to pick from, a draft to approve or edit, or a question) and send your answer. The assistant never answers a checkpoint for you, and gives you a link to answer it in the dashboard instead if you prefer
  • Run it again with changed inputs or answers. The assistant first shows which stages would run again, which results are reused, and what it would cost
  • Explain why a run stopped, or cancel a run that is still going
Try asking: “Run my article pipeline for the keyword running shoes and show me the outline before it writes.”

Approvals

  • Show app actions that are waiting for a person to review
  • Allow an action once or deny it, after you tell the assistant your answer
To always allow an action, the assistant gives you a link to the review page in the dashboard.