October 22, 2026 - AI agents upgraded to Vercel AI SDK 7 with stricter API validation

Change Log Application Enablement
, ,
1

Context

Change Type: Feature
Product area: Application enablement & solutions
Component: AI Agents
Deployed at: eu.latest.cumulocity.com, apj.cumulocity.com, jp.cumulocity.com, cumulocity.com, us.cumulocity.com

Technical details

Build artifact: ai-agents (3.0.0)
Internal ID: MTM-67854

Description

The AI Agents microservice now uses version 7 of the Vercel AI SDK. You can now define an agent’s system prompt as a single system message or as a list of system messages. This lets you set provider-specific options on the system prompt, for example, to enable Anthropic prompt caching and reduce cost and latency for repeated requests.

Important: This release contains breaking changes to the agent API. If you call agents directly through the API, review the following changes and update your integrations accordingly:

  • Requests to POST /agent/text/{name} and POST /agent/object/{name} now accept only the messages, prompt and variables fields. Requests that contain any other field are rejected with HTTP status 400.
  • The messages field can no longer contain messages with role: "system". Define the system prompt in the system field of the agent definition instead.
  • Full responses (fullResponse=true) no longer include the raw request and response bodies of each step (steps[].request.body and steps[].response.body).
  • In streamed responses, start-step events no longer include request.body. To identify the model that was used, read response.modelId from the finish-step event.
  • The output of object agents is now validated against the agent’s JSON schema, including string formats such as date-time or email. If the output does not match the schema, the request fails with HTTP status 400 instead of returning the invalid output.