Why Does the JSONB Metadata Filter Fail in n8n’s AI Agent Mode with Supabase?

Below is some background that might help troubleshoot the issue:

Understanding the Modes
The Supabase Vector Store node for n8n offers several modes. In Get Many mode—as used with “Retrieve documents as vector store for chain tools”—the metadata filter is supported as an additional AND query. However, when using the node with the “Retrieve documents as tool for AI Agent” mode, the node internally applies its own query parameters. This can lead to a conflict if you also pass a metadata filter manually.

What to Check
• Ensure that your metadata filter value is formatted exactly as expected (key-value pairs) based on the documentation. Even a slight mismatch in JSONB formatting might cause the filter to fail.
• Because the AI Agent mode may combine built-in metadata filtering with your provided filter, verify that only one source of filtering is applied. In other words, if the agent mode already applies a filter, you might need to remove or adjust the manual filter.
• Review your logic: if you need complex filtering, consider switching to the chain tools mode where the metadata filter is fully supported.

Next Steps

1. Try executing the query in the agent mode without the manual metadata filter. Confirm if that works without error.
2. If filtering is required for your agent mode execution, double-check the JSONB structure and ensure that no additional (hidden) filter is applied by the node that could conflict with it.
3. Consult the Supabase quickstart or related documentation for examples of metadata filters working with JSONB in a similar context. Sometimes slight differences in parameter handling (for example, extra filters coming from the node’s defaults) can disrupt the intended behavior.

If after these steps the issue persists, please provide a minimal reproducible example of your node configuration (without sensitive information) so that further troubleshooting can be done.

Hope this helps clarify the filtering conflict!