Overview
Teams can now give more people visibility into AI Agents without granting Pipe admin access. Admins can turn on, per Pipe, view only access to Agents and their execution history, without granting permission to create, edit or delete Agents.
What's new?
- Permission configurable per Pipe, under Pipe Settings › AI Features
- View only access to AI Agents: members can inspect an Agent's configuration without permission to change it
- Access to Agent Logs: members can review execution history, including prompts, field values, knowledge base chunks and raw outputs
- Off by default: access is only granted when a Pipe Admin explicitly enables it, and it applies only to that Pipe
How does this help?
- More visibility, less risk: more people can diagnose an Agent's behavior without expanding admin access on the Pipe
- Faster troubleshooting: whoever needs to investigate an execution sees prompts, fields and outputs right in the Pipe, without depending on an Admin for every lookup
- Protected configuration: the Agent stays protected against accidental changes, since view only access does not include editing
Use cases
Operations Analyst investigates an automation failure. Without needing Admin access on the Pipe, the analyst responsible for daily ticket triage checks the Agent Log and finds that a field came through empty in the prompt, causing an unexpected response. With the cause mapped, the fix request goes straight to the Pipe Admin, no screenshots or manual descriptions of the execution needed.
Internal Support team validates a service Agent's answers. A member of the support Pipe reviews, in the Logs, the knowledge base chunks the Agent used to answer customer questions about refund policy. Noticing the Agent pulled an outdated version of the document, the team triggers a content update without needing access to edit the Agent's configuration.
New team member follows an Agent without risk of changing its setup. During onboarding, a new member of the financial approval Pipe gets view only access to understand how the Agent classifies requests by value and urgency. They follow real executions without permission to create, edit or delete the Agent, reducing the risk of a configuration being changed by mistake while learning.
Pipe Admin delegates investigation of a specific incident. When an IDP Agent starts extracting incorrect values from a specific document type, the Pipe Admin enables view only access for the technical lead to investigate. That person checks the execution's field values and raw outputs, finds the cause and reports back, without ever having permission to change the Agent.
Most relevant for
- Pipe Admins
- Operations teams
- AI Agent builders
- Internal Support teams
Plans and availability
This feature is being rolled out gradually.
