# Release 2026.6
Check June's updates.
***
## New Models: Claude 5 Sonnet, Claude 5 Fable, and More
Several new models are now available in the model selector.
### Claude 5 Sonnet and Claude 5 Fable
Claude 5 Sonnet and Claude 5 Fable are now available as models through both the Anthropic and AWS Bedrock providers. Both models support up to 128,000 output tokens and include an *Adaptive Thinking Effort* parameter (`adaptive_thinking_effort`) that lets you control reasoning depth. Set it to `low`, `medium`, `high`, `xhigh`, or `max`.
Claude 5 Fable on AWS Bedrock is available in the `eu-west-1` region.
### Updated Token Limits for Claude 4.6 and 4.7
The maximum output tokens for Claude 4.6 and Claude 4.7 models have been increased from 64,000 to 128,000 tokens.
### New Gemini Models
Gemini 3.5 Flash and Gemini 3.1 Flash Light are now available. Both include a `thinking_level` parameter with options `minimal`, `low`, `medium`, and `high`. Gemini 3.1 Pro now supports a `medium` thinking level in addition to `low` and `high`.
For information on using models in your pipelines, see [Language Models in Haystack Enterprise Platform](/docs/concepts/language-models-in-deepset-cloud.mdx).
## Image Preview Now Available in Files and Indexes
You can now preview image files directly in the Files and Indexes pages. Click **View** next to any PNG, JPG, JPEG, GIF, WEBP, or SVG file to open it in the file viewer. Previously, image files did not show a **View** button even though the preview infrastructure was in place.
For the full list of supported preview formats, see [Upload Files](/docs/how-to-guides/working-with-your-data/upload-files.mdx).
## Anthropic on Azure Now Available as a Custom Model Provider
You can now use Anthropic on Azure as a provider when adding a custom model definition. Select **Anthropic on Azure** from the provider list in *Settings > Integrations > Custom Models* to configure models hosted on Azure through Anthropic Foundry.
For details on setting up custom models, see [Add Custom Model Definitions](/docs/how-to-guides/managing-access/add-custom-model-definitions.mdx).
## Use Your Pipelines as MCP Tools
You can now use your deployed pipelines as MCP tools, making them available to AI coding assistants like Cursor, VS Code, or Claude Code. In , this works at the workspace level: you create one MCP server for your workspace, then individually enable each pipeline as its tool. A pipeline is not available as a tool until you explicitly turn it on in the pipeline's settings.
When an AI client connects to your workspace's MCP endpoint, it discovers all the pipelines you enabled and can call them as tools. Each tool can have its own name, description, and instructions to help the AI assistant understand when and how to use it.
This means you can mix and match pipelines within a single MCP server — for example, a RAG pipeline for knowledge retrieval, a data processing pipeline for transformation tasks, and a summarization pipeline for document analysis — all accessible through one endpoint.
For details, see [Use Your Pipelines as MCP Tools](/docs/how-to-guides/productionizing-your-pipeline/expose-pipeline-as-mcp-tool.mdx).
## Mem0 Memory Tools for Agents
You can now give your Agent long-term memory using Mem0. Adding Mem0 Memory to an Agent creates two tools: one that stores memories and one that retrieves them. Both tools are scoped to individual users so each user's memory stays separate.
The *store memory* tool lets the Agent save facts, preferences, and context from a conversation. The *retrieve memories* tool lets the Agent search those stored memories when answering the user.
To add Mem0 Memory tools, open the Agent configuration panel, click **Add Tool**, and choose *Mem0 Memory*. You need a Mem0 API key, which is saved as a workspace secret.
For details, see [Agent Tools](/docs/concepts/ai-agents/agent-tools.mdx#mem0-memory) and [Add Mem0 Memory Tools](/docs/how-to-guides/building-agents/configuring-agent.mdx#add-mem0-memory-tools).
### Save Pipeline Tools Despite Validation Errors
You can now save a pipeline tool even when the underlying pipeline has validation errors, such as an embedding dimension mismatch or a schema error. Previously, pipeline tool creation was blocked in these cases. Now, validation messages remain visible so you can review and fix them, but they no longer prevent you from saving the tool.
The tool validation response also includes a new `can_create` field that signals whether creation is allowed. For pipeline tools, `can_create` is `true` whenever a valid tool definition exists — even if the pipeline has errors. For MCP and other tool types, `can_create` is `true` only when there are no validation errors.
For details, see [Configure an Agent](/docs/how-to-guides/building-agents/configuring-agent.mdx).
## Builder Updates
### New Layout
We've revamped Builder to make the building experience complete. Now, you can switch between Builder, Playground, pipeline analytics, and settings all in one place. No more switching between pages.
### Simplified Component Layouts and Clearer Connections
Building pipelines just got cleaner. We've redesigned component layouts to give you a more transparent view of your pipeline. Now, you configure components in a configuration panel that opens when you click the component card.
We've also added an expandable Connections panel to help you understand how components are connected and how data flows through your pipeline. Expand it for a full overview, or click any component to inspect its connections up close.
You can now also use the Command Palette to create connections. Press **CMD + K** to open it and type `connect` to see the available commands.
### Configure Code Component Init Parameters from the UI
You can add `__init__` parameters to your `Code` components. Just make sure they have default values. When you add `__init__` parameters, they're shown on the Advanced tab on the `Code` component's configuration. String, number, boolean, and enum parameters each get a dedicated inline control. Complex types such as objects or dictionaries open in an expandable editor. Values you set on the **Advanced** tab are saved into the pipeline configuration automatically.
For details, see [Add a Code Component](/docs/how-to-guides/designing-your-pipeline/work-with-custom-components/create-code-component.mdx).
### Global Issues Panel
Builder now has a global Issues panel at the bottom of the canvas, next to the Connections tab. It collects all validation and runtime issues in one place, so you don't have to hunt through individual components to find out what's wrong.
When an issue appears, click **Inspect** next to it to jump directly to the affected component or connection and fix it.
### MultiRetriever Knowledge Source Improvements
The **Add/Edit Knowledge Source** flow in the MultiRetriever component has been updated with several improvements. You can now connect external document stores such as Pinecone and Qdrant by typing the index name directly and entering connection credentials. Credential fields support workspace and organization secrets — select a saved secret from the dropdown to fill the field securely without exposing raw values. Retriever parameters are now fully exposed in the configuration panel, and knowledge source cards show a yellow warning indicator when required fields are not yet filled in.
For details, see [MultiRetriever](/docs/reference/pipeline-components/knowledge-retrieval/MultiRetriever.mdx).
### Model Picker in Component Drawers
Components that use an embedded chat generator — such as `LLMMetadataExtractor` — now show a model picker in their configuration drawer instead of a raw YAML editor. Select a model from the dropdown, use the **Connected only** toggle to filter to models with active provider connections, and get connection alerts directly in the drawer if the selected model is not connected. This matches the model selection experience already available in `LLM` and `Agent` component drawers.
This release also fixes false **Model not connected** warnings that previously appeared in the Issues panel for these components even when the provider was connected.
## Type Literal `@` and `/` in the Prompt Editor
You can now type a literal `@` or `/` character in the Jinja2 prompt editor when configuring the prompt for the `LLM` or `Agent` components. Typing either character opens a list of available variables or functions, but if you press `Escape` after the list appears, the character is inserted into the prompt.
## More Control Over Pipeline Versions
Two new actions are now available in the pipeline versions list in Builder:
- **Finalize** — converts a draft version into a permanent, numbered version. Use this when you want to lock in a draft without deploying it.
- **Set as default** — marks a non-draft version as the default. The default version is used by serverless inference endpoints when a request identifies a pipeline by name or ID but does not specify a version. Default versions appear with a gold star **Default** tag in the versions list.
For details, see [Manage Pipeline Versions](/docs/how-to-guides/designing-your-pipeline/manage-pipeline-versions.mdx).
## Search History Updates
### Notes and Hashtags on Search History Records
Add text notes to individual search history records to help you keep track of important information. Use notes to record observations, flag interesting queries, or leave context for your team. Notes are shown in the Search History table on the Pipeline Details page and in the detailed view for each query. You can add, edit, and delete notes at any time.
You can also add hashtags to group related records together.
Both notes and hashtags are searchable. You can view and manage them if you enable the Notes and Hashtags columns in the Search History table.
## Validation Warnings No Longer Block Pipeline Deployment
Pipeline validation messages are now split into two categories: errors and warnings. Only errors prevent a pipeline from being deployed. Warnings are informational hints about your configuration and do not block deployment — your pipeline deploys successfully even when warnings are present. Review warnings to make sure your configuration is intentional.
For more information, see [Troubleshoot Pipelines and Indexes](/docs/how-to-guides/designing-your-pipeline/troubleshooting-pipeline-deployment.mdx).
### Validation Warnings No Longer Block Pipeline Tool Creation
This same behavior now applies when adding a pipeline as an agent tool. If the pipeline you're adding has validation warnings but no hard errors, the tool is created successfully. Any warnings are stored on the tool and visible in the Agent Settings panel so you can review them. Only hard failures — such as a pipeline that cannot be parsed — block tool creation.
For details on adding pipeline tools, see [Configure an Agent](/docs/how-to-guides/building-agents/configuring-agent.mdx).
### AI-Powered Feedback insights
Use the AI-powered Feedback Insights panel to analyze feedback patterns and generate executive summaries to help you understand performance trends and identify areas for improvement.
Click Search History on the Pipeline Details page, expand the Feedback Insights panel, and use one of the suggested prompts or your own query to get insights.
## Validation Warnings No Longer Block Pipeline Deployment
Pipeline validation messages are now split into two categories: errors and warnings. Only errors prevent a pipeline from being deployed. Warnings are informational hints about your configuration and do not block deployment — your pipeline deploys successfully even when warnings are present. Review warnings to make sure your configuration is intentional.
For more information, see [Troubleshoot Pipelines and Indexes](/docs/how-to-guides/designing-your-pipeline/troubleshooting-pipeline-deployment.mdx).
## Editor Role Can Now Manage Workspace Custom Models
The Editor workspace role now has write access to workspace-scoped custom model definitions. Editors can create, update, and delete custom models in their workspace without needing an Admin role. Organization-level custom models remain Admin-only.
For details, see [User Roles and Permissions](/docs/concepts/user-roles-and-permissions.mdx) and [Add Custom Model Definitions](/docs/how-to-guides/managing-access/add-custom-model-definitions.mdx).