A: Hey Steve_P!

That's a fantastic question and absolutely key for keeping agents current and effective.

Yes, there are ways to update agent knowledge in an automated or semi-automated fashion, and connecting to data sources like Airtable is definitely part of the vision and current capability.

Here's how you can approach it:
- Updating Knowledge from Uploaded Documents (More Manual for Direct Agent KB currently):
- When you directly upload files (PDFs, DOCX, etc.) into an Agent's specific Knowledge Base using the new Data section feature, updating that knowledge currently involves re-uploading the new or modified file.
- Automation Potential (via API): We are working on exposing API endpoints to manage an Agent's direct Knowledge Base. Once live, you could theoretically build a Make.com scenario that watches for new/updated files (e.g., in a Google Drive folder) and then uses an HTTP module in Make to call our API to update the agent's knowledge with the new file content. This is on our development path.

Connecting Agents to Dynamic Data Sources (like Airtable - Recommended for Automated Updates):
- This is often the more robust and flexible way for knowledge that changes frequently.
- Airtable MCP: Connect your Airtable base using our Airtable MCP (https://agenticflow.ai/mcp/airtable_oauth).

How it Works:
Instead of (or in addition to) uploading static docs, you design your agent to query Airtable live when it needs information.
Agent's System Prompt / Workflow Logic: You instruct the agent: "When a user asks about [X], search the 'Product Specs' table in our Airtable base for relevant information using the Airtable tool, then use that to answer."
No "Retraining" Needed for Airtable Data Changes: If you update a record in your Airtable base, the next time the agent is instructed to query that table, it will automatically fetch the latest, updated information from Airtable directly. There's no separate "retraining" or "knowledge update" step needed within AgenticFlow for data residing in Airtable itself. The knowledge is always "live."

Using Airtable as a Knowledge Source for RAG:

You can also create an AgenticFlow Table Dataset and populate it by regularly pulling data from your Airtable base (e.g., a scheduled workflow that uses the Airtable MCP to get records and then updates the AgenticFlow Table Dataset).

Then, enable "Knowledge" on this AgenticFlow Table Dataset. The agent will then use this internal, vectorized copy for its RAG searches. This requires periodic updates of the AgenticFlow Table Dataset from Airtable.

Using Make.com (or similar) for Updates:

Scenario 1 (Updating an AgenticFlow Table Dataset from Airtable via Make):
- Make.com: Trigger on "New/Updated Record in Airtable."
- Make.com: Use an HTTP module to call the AgenticFlow API to "Update Row" or "Add Row" in a specific AgenticFlow Table Dataset that your agent uses for knowledge.

Scenario 2 (Triggering AgenticFlow to pull from Airtable via Make):
- Make.com: Trigger on a schedule or event.
- Make.com: Use an HTTP module to call an AgenticFlow workflow's API endpoint.

AgenticFlow Workflow: This workflow then uses the Airtable MCP to fetch the latest data and either updates an internal Table Dataset or directly uses it for a task.

In Short:
- For static docs directly uploaded to an agent, updates are manual for now (API for this coming).
- For dynamic knowledge, connect your agent to live databases like Airtable via MCP. The agent queries the live data, so updates in Airtable are reflected instantly without needing to "retrain" the agent on that specific data point.
- You can use Make.com to automate syncing data from Airtable into an AgenticFlow Table Dataset if you prefer to use our internal RAG on a snapshot of the data.

The "live query via MCP" to Airtable is often the most straightforward for automatically current knowledge.
— Sean