3.11.3 Release Notes#

Squirro 3.11.3 was released on October 23, 2024.

Note for administrators

This release introduces important changes that may require adjustments to your existing setup.

What’s New#

Added a new Llama schema for SquirroGPT configurations.
Launched the Incident Support Agent as part of the multi-agent functionality.
Support for two new LLM providers, Google AI and Cerebras.
Introduced a generic prompt API for the GenAI service.
Implemented autocompletion support for the user document collections using the collection suggester.
The Paragraph Embedding step now includes an advanced option for configuring vector storage in Elasticsearch. It allows the simultaneous storage of float and byte vectors for enhanced precision and options for brute force search instead of building an HNSW graph.
The Paragraph Embedding step now includes a timeout option for the requests to the sentence embeddings service, configurable at both the step and server levels (using the topic.nlp.remote-services-connection option), with the step-level settings taking precedence. It ensures that no request to the sentence embeddings service waits indefinitely, preventing ingestion delays.
Introduced new CryptKeeper``classes that resolve configuration values through environment variables using the ``CK_ENV::<value> syntax. For instance, CK_ENV::EXAMPLE_ENV fetches the value of the EXAMPLE_ENV environment variable.
Introduced the new React Rich Text Editor widget.
The Data Ingestion Logs dashboard now displays project titles and data source names and allows filtering by both.

When using the queue-based filesystem streamer (QFSS), the source queue statistics from Redis are now deleted upon data source deletion. In addition, when the index of queues is built, the queue statistics are updated.
Ensured application of the project visual theme before the page loads, eliminating the transient display of non-themed content.
Enhanced the GenAI RPM package to include the Docker images as .tar.gz for offline installations.
Enabled intent_detector to accept the label_lookup_rescore_parameters parameter for customized label rescoring, for example, adjusting rescore_query_weight to prioritize rescored results.
Added support for *.ppsx and *.potx files (PowerPoint slideshows and templates).
Enhanced structured data ingestion logs to include the project title and data source name as labels.
Introduced paragraph-scoped key phrase and NER tagging. Previously, all NLP tags were inherited from the parent item.
Forwarding of all payload arguments to model_extras, allowing access to chat_history within a tool.
Improved context_guessing functionality for unnamed models.
Improved keyword sorting by display value in item views.
Automatic expansion option for the time selection widget.

Respect the MAX_FILES setting in the file upload control for the Data Loader plugin.
Fixed the duplicated Crawler entry in the Web tab and removed redundant entries for the Dropbox and Google Drive connectors from the Enterprise tab.
The streaming chat endpoint now overwrites the payload tokens with the query parameter tokens when both are provided.
Removed a misleading warning log message from FileSystemStreamer.get_number_of_batches logged inside ingester.log.
Fixed issues with dynamic template overwriting during migrations, ensuring proper appending.
Resolved highlighting issues in items using the FuzzyMatch plugin in the in_order mode.
Fixed 403 errors occurring on websites with empty Authorization headers.
Collections filter now compatible with the semantic search.
Resolved issues causing balanced AutoML templates to fail during model training.
Ensured project deletions clear out project-specific caches as expected.
Removed epoch hyperparameter from the AutoGluon classifier due to its deprecation, allowing automatic adjustment of epochs instead of having a fixated value.
Corrected Redis instance usage by QFSS for Ansible deployments, to use redis-server instead of redis-server-cache.

Reverted gunicorn introduction for topic and frontend services due to unstable behaviors.
The GenAI service RPM is now a weak dependency of the clusternode service RPM for RHEL 8. By default, package managers should install squirro.service.genai, but installation failures will not affect teh squirro-cluster-node installation. Use install_weak_deps=False for dnf or yum to turn off the installation. The squirro.service.genai package can be independently uninstalled without affecting squirro-cluster-node to support deployments where they run on different machines.
The top_k_aggregation function now requires explicit specifying communities in fields.

For new installations, find step-by-step instructions on the Install and Manage Squirro with Ansible page (recommended) and Installing Squirro on Linux pages.

To upgrade an existing installation, see the Upgrading Squirro page.