Release 2026.9
Catch Pipeline Dependency Issues Earlier
Haystack Enterprise Platform now checks your pipeline YAML dependencies block when you validate or deploy a pipeline. If a dependency isn't approved, uses a direct URL reference, or isn't a valid requirement string, validation points to the specific entry so you can fix it before the pipeline environment is built.
For details, see Pipeline Dependencies. To resolve dependency validation errors, see Troubleshoot Pipelines and Indexes.
Session Storage Pipelines Use Chat Output
Pipelines with session_storage: true at the top level of their YAML now automatically use the chat output type in Playground and the API, even if they don't contain generators or chat-related prompts. Session storage needs a session ID, and the chat interface provides one.
To use a different output type, set pipeline_output_type explicitly in your pipeline YAML. For details, see Set Pipeline Output Type and Update Pipeline Settings.
Smoother Trace Browsing
The Analytics > Traces view got easier to use: export a trace as JSON directly from its row menu without opening it first, jump straight to the History tab with your query preloaded using View in History, and reset the source filter with the new All option. See Trace with Built-In Traces for details.
Easier Debugging of Agent Tools
Testing tools in Builder just got clearer:
- Run Tool form now shows every input your tool needs — not just the ones marked required, so you won't hit missing-parameter errors you can't explain.
- Error messages point to the real problem. Instead of a generic wrapper message, you'll see the exact exception, like
ValueError: Missing mandatory input 'query' for component 'SerperDevWebSearch'.
Credits Used on Job Details
Organization admins can now see how many credits a job used directly on its Job Details page, shown next to Duration once the job's usage is a positive number. This makes it easier to spot which jobs are driving usage without leaving the job's page. For more information, see Jobs.
Smarter AI Assistant
Code component, Code Tool, and Conditional Router now open the same AI Assistant to help you generate code or routing logic. The component or tool you're working on stays attached as context, so you can keep asking follow-up questions without losing track of what you're editing. Attached context now works alongside your pipeline's baseline info, so the assistant has the full picture when it responds.
Click the AI Assistant icon in the header to open it.
To learn how to use these surfaces, see Add a Code Component, Agent Tools, or ConditionalRouter.
Was this page helpful?