Overview
Uploading files to an agent is a one-time act. The world it answers questions about is not. A price list changes, a policy page gets rewritten, new questions arrive from a form, and an agent trained last month starts giving last month's answers with total confidence.
Your agent's Knowledge area has an Automation tab that fixes this. Point it at something that changes, and Taskade feeds the new version in for you.
TL;DR: Open your agent's Knowledge → Automation tab and pick a recipe. Your agent can watch a web page, sync a Google Sheet or Airtable, learn from forwarded email, absorb new form replies, or capture key Slack and Discord messages. Set it once and answers stay current. For the one-time upload path, see Agent Knowledge & Memory.
Where the Automation Tab Lives
Open any custom agent and go to Knowledge. Three tabs sit across the top:
┌──────────────────────────────────────────────────────────────┐
│ Edit AI Agent › Knowledge │
├──────────────────────────────────────────────────────────────┤
│ Instructions │ Sources │ [ Automation ] │
├──────────────────────────────────────────────────────────────┤
│ │
│ Scheduled updates │
│ On a schedule, pull fresh content into your agent │
│ knowledge. │
│ │
│ Email to knowledge │
│ Forward an email and your agent learns from it │
│ automatically. │
│ │
│ Form responses │
│ Turn new form submissions into agent knowledge as they │
│ arrive. │
│ │
│ Chat intelligence │
│ Capture key messages from Slack or Discord into your │
│ agent. │
│ │
│ Spreadsheet sync │
│ Keep your agent in sync with a Google Sheet or Airtable. │
│ │
│ Web scraping │
│ Watch a web page and train your agent when it changes. │
│ │
└──────────────────────────────────────────────────────────────┘
Instructions shapes how the agent answers. Sources holds what you uploaded by hand. Automation is the part that keeps that material fresh on its own.
How a Refresh Works
Each recipe is a small automation with your agent's knowledge as its destination. Something changes out in the world, the flow runs, and the new material lands in the agent's knowledge before anyone asks a question about it.
Nobody has to remember to re-upload a file. That is the whole point.
The Six Recipes
| Recipe | What it watches | Reach for it when |
|---|---|---|
| Scheduled updates | Any content you point it at, on a repeating schedule | You want a standing refresh rather than a trigger |
| Web scraping | A single web page, retraining when it changes | Pricing, policy, or docs pages that get edited |
| Spreadsheet sync | A Google Sheet or Airtable base | Inventory, rosters, rates, or anything living in a sheet |
| Form responses | New submissions to a form | Support questions and intake replies as they arrive |
| Email to knowledge | Messages you forward to it | Supplier notices and updates that only arrive by email |
| Chat intelligence | Key messages in Slack or Discord | Decisions and answers that only ever get said in chat |
Pick the One That Matches Your Source
The useful question is not "which is best" but "where does the truth already live?" Point the recipe at that place rather than making someone copy it somewhere else first.
- The answer lives on a page on your website → Web scraping
- The answer lives in a spreadsheet someone already maintains → Spreadsheet sync
- The answer arrives as a message → Email to knowledge, or Chat intelligence for Slack and Discord
- The answer arrives as a form reply → Form responses
- The answer changes on a predictable rhythm → Scheduled updates
A good habit: pick the source your team already updates for their own reasons. A sheet the ops team edits weekly will stay accurate on its own. A document created only to feed the agent will quietly go stale.
Turn One On
- Open your agent and go to Knowledge.
- Select the Automation tab.
- Choose the recipe that matches where your material lives.
- Connect the account or page it needs, and set how often it runs.
- Ask the agent a question whose answer only appears in the new material. If it answers correctly, the loop is closed.
That last step is the one people skip. It is also the only real proof the wiring works.
One quirk worth knowing: the Knowledge screen does not update itself while you sit on it. When a recipe runs and adds something, reopen the screen or refresh the page to see the new entry in the list. Do not read a short list as a failed run, and do not wait on the list at all. Asking the agent is the faster and better check, which is exactly why it is step 5.
Most of these recipes lean on premium connectors such as Google Sheets, Gmail, and Slack, and those steps need a paid plan. A schedule-driven recipe built only on the always-available steps runs on Free, which includes 10 automation runs a month. Everything on the Sources tab, including uploading files by hand, works on every plan.
Keeping It Trustworthy
An agent that refreshes itself can also absorb something you did not intend, so a little care pays off:
- Point recipes at sources you control. A page you own is a promise. A page you do not own can change in ways that surprise you.
- Prefer one clear source per fact. When two sources disagree, an agent will happily quote either one.
- Check the citations in an answer to see which source it actually used. Agent Knowledge & Memory covers how sources appear in replies.
- Remove sources that no longer matter. Stale material does not become harmless because newer material exists alongside it.
Related guides
- Agent Knowledge & Memory. Upload files, add web resources, and connect live projects by hand
- Build a Chatbot with Knowledge Base. The full path from documents to a published website widget
- Custom AI Agents. How agents are built and configured
- Automations. The execution layer these recipes run on
- Agent Tools. Let an agent act, not only answer