Foundry Agents Not Working

Hello Platform,

Welcome to Microsoft Q&A .Thank you for reaching out to us.

The behavior aligns with a known pattern of backend latency or dependency-related degradation, rather than a confirmed global outage. The issue is most commonly observed in specific regions or service layers such as vector store processing.

Recent observations indicate that the current behavior experienced with Foundry Agents is related to a service‑side dependency issue rather than an isolated configuration concern. The errors reported—such as failures when opening agents in the Agent Playground, inability to update agent settings, unsuccessful file attachments, and timeouts while fetching vector stores—are all interconnected and stem from the same underlying execution path.

When an agent is opened or edited, the platform retrieves agent metadata, connected tools, and any associated vector stores used for file grounding. If one of these backend services becomes slow or temporarily unavailable, the request exceeds the timeout threshold and results in messages such as *“*Service temporarily unavailable” or “Unable to fetch vector stores – Timeout of 60000ms exceeded.” Since file uploads and agent edits rely on this same control‑plane orchestration, all related actions fail together instead of independently.

This behavior has been recently reported by multiple tenants and is currently under investigation by engineering teams. While global service health dashboards may continue to appear operational, these dashboards do not always surface tenant‑specific or region‑specific orchestration degradations, especially for preview‑stage experiences such as Foundry Agents.

Meanwhile , please check if the following steps help -

1. Isolating vector store dependency
   • First create a temporary agent without file grounding enabled.
   • Then confirm whether basic prompt execution succeeds.
   • A successful execution indicates vector store orchestration latency rather than core agent failure.
2. Validating outside the portal
   • Invoke the existing agent using SDKs or REST APIs.
   • This helps distinguish portal‑level control‑plane impact from runtime inference availability.
3. Confirming regional behavior
   • If feasible, please test agent creation or invocation in an alternate supported region.
   • This assists in identifying region‑scoped service degradation.
4. Reviewing dependent services
   • Ensure connected storage, search, and identity services remain accessible.
   • No changes are required unless recent modifications were made prior to the issue.
5. Performing session‑level checks
   • Test access using a private browser session or an alternate browser.
   • This eliminates cached portal state as a contributing factor.

Current indications show no loss of agent data. Once backend service stability is restored, agent editing, file uploads, and vector store access typically resume without requiring reconfiguration.

We appreciate your patience while the respective teams are working on the resolution.

Thank you.

Similar problem here.

Using the TRY IN PLAYGROUND option to test my agent:

Failed to create thread Service temporarily unavailable. Please try again later Trace ID : <REDACTED>Client request ID : REDACTED APIM request ID : REDACTED.

Note: PII Redacted at support side.

I have this same error as well since at least several hours ago, potentially yesterday too.

Hello Platform,

Checking in to see if you had any chance to review the above response.

Do let me know if you have any further queries.

Thank you