Siren AI changelog

2.0.0

• Breaking changes
  • The siren-ai configuration format has changed to support multiple models. Please refer to the documentation for the updated configuration format
• New features
  • K9 now supports multiple languages with full i18n support, enabling global usage
  • Added tool support for AWS Bedrock enabling K9 to perform actions when using AWS-hosted models
  • Ollama now uses a generic OpenAI-compatible provider config instead of Ollama-specific one, improving compatibility and flexibility
  • Users can now select from multiple configured AI models at runtime
  • Edge references now render as interactive badges with visual styling, improving graph readability
  • Added keyboard shortcuts to quickly open and close K9 interface without using mouse
  • Added visible close button for K9 popover instead of relying only on clicking outside
  • K9 now has awareness of and can reference combo/group nodes in graphs, enhancing support for complex graph structures
• Improvements
  • K9 global search tool now supports non-range date queries, providing more flexible filtering options
  • AI provider errors are now displayed directly within K9, making troubleshooting easier
  • Added retry mechanism for responses containing rate limit messages, improving resilience
  • K9 popover can now be resized by users for better control over interface layout
  • K9 has knowledge of visible and hidden nodes and edges in the graph
  • Added node selection action badges for improved interaction feedback
  • Chat window now scrolls to the top of new K9 messages
  • Fixed issue where badges can’t select records from indices that have certain characters like :
  • Fixed graph serialization issues for nodes with array fields
  • Fixed error when Siren AI plugin is disabled via configuration

1.4.0

• New features
  • K9 can now perform global searches, searching relevant entity tables with the appropriate dynamic filters
    Clicking on the search action in the conversation window allows you to review the parameters K9 used and the results it got back
  • K9 can create new graphs from search results and other data
  • K9 can add nodes, select nodes and group/ungroup nodes in graphs
  • K9 can add nodes to unopened saved graphs
  • K9 can see results from searches you make in the Search app and use them to perform actions (e.g. creating a graph)
  • Ollama can now use tools to perform actions like Azure and OpenAI
• Improvements
  • You can now define a timeout for LLMs in the Investigate server config
  • When K9 fails to perform an action, the error message can be seen in the conversation history
  • K9 can understand grouped edges in the graph
  • K9 can now see record data used in record-as-relation edges
  • Graph reports now display clickable badges when referencing records to help you find them in the graph
  • Updated default user prompt in advanced settings to improve K9’s assistive behaviour
  • Various bug fixes and robustness improvements
  • K9 now wags his tail when generating a response.

1.3.3

• Fixed an issue where K9 could not see selected nodes in the serialized graph
• Fixed an issue where K9 could not see grouped edges in the graph

1.3.2

• Fixed issue where K9 could not see records-as-relations from dataspaces other than HOME
• Minor documentation improvements

1.3.1

• Further optimization to the amount of graph data sent to the LLM
• Fixed incorrect graph edge data being sent to the LLM
• Fixed aggregated relations not being seen by the LLM
• Added doc section describing what graph data is used
• Fixed K9 chat window not scrolling to the bottom when new messages are sent/received on Firefox

1.3.0

• K9 chat and report generation prompts can now be customized in the advanced settings
• Graph serialization has been turbo-charged, so now graphs up to three times larger can be sent to the LLM
• Minor doc improvements
• Fixed issue where references to EIDs in the graph aren’t clickable
• Fixed issue where EIDs couldn’t be styled by K9 Chat Companion

1.2.0

• K9 improvements
  • References to documents in K9’s messages are now clickable and will open in the record viewer
  • Support for styling edges in the graph
  • Support for local nodes in graphs
  • Emoticons are no longer converted to emojis :)
• Minor UI improvements
• Improved error messages
  • When graph is too large
  • When generated graph report can not be downloaded
• Documentation fixes
  • Fixed some scripting examples

1.1.0

• K9 Companion now available!
  • Chat with the LLM from the Investigate UI
  • Graph-aware: You can ask questions about the graph you’re looking at
  • K9 can set node styles in response to your questions
• Improvements to generated graph report
  • Added an image of the graph to downloaded reports
  • Report title now uses the name of the graph
  • “Introduction” section removed
  • Renamed “Graph Nodes” section to “Selected Nodes”
• Scripting API improvements:
  • Support for a custom LLM prompt when generating a report
  • Added autocompletion for Siren AI methods in the Investigate script editor
• Other minor UX improvements
• Improved error messaging
  • When Siren AI plugin is disabled
  • When graph is too big
• Documentation fixes:
  • Updated config option “ollama.connectin.endpoint” to “ollama.connection.host”

1.0.1

• Fix crash when using Investigate with security disabled

1.0.0

• Generate downoadable reports using AI from the graph browser
• Siren scripting API methods for sending messages and generating reports
• Support for OpenAI, Azure OpenAI, Ollama and AWS Bedrock
• Permissions for configuraing access to Siren AI features
• Support for auditing LLM queries