Assistant
The Assistant tab is a place where you can test your active graph models. This feature allows users to chat with their model and receive responses based on knowledge from them.
How the Assistant works?
The Assistant searches only within analyzed data to provide relevant answers in a conversational format.
It does not use external knowledge sources.
Responses are based strictly on retrieved results.
If no relevant data is found, the system will inform about it.
The chat interface is powered by an external AWS Bedrock LLM. It is used only to present Galaxia’s knowledge-retrieval results in a more human-sounding way. Chat behavior and limitations remain independent of Galaxia’s deterministic reasoning.
Creating a Conversation
To start a new conversation:
Click Create.
Select a Model from the dropdown list.
Review the automatically assigned Model name.
(Optional) Modify the default prompt. It must be shorter than 4000 characters.
Click Create to open the conversation.
Additionally, if the model remains inactive for 30 minutes, it enters a sleep mode. When this happens, the first query after inactivity may take longer to process, as it wakes the model up. The wake-up process is triggered by sending a new query.
Assistant interface
The Assistant interface consists of three main areas:
1. Conversation list (left panel)
Displays all created conversations.
Use the Search field to find a specific conversation.
Click on a conversation name to open it.
You can maintain multiple conversations for different models or use cases.
Hover over a conversation to show the delete icon on the right and remove it.
2. Chat window (center panel)
This is where you interact with the model.
Enter your question or query in the input field at the bottom.
Press Enter or click the send icon to submit.
The model responds using data retrieved from your built graph model.
If the system cannot find sufficient information, it returns a message indicating that there is not enough data to answer the query.
The chat maintains context from the last 5 queries, allowing for follow-up questions.
The current text input limit is 200 characters.
Current version is optimized for English language queries.
3. Query results (right panel)
For each question, detailed retrieval results can be displayed by clicking details
This panel includes:
Query – the searched keyword or phrase.
Results – matched records from the dataset.
Matches – matched terms.
Confidence – relevance level of the match.
See in graph – option to visualize the result in the graph view.
If multiple queries are automatically issued (for example, when the initial query does not return sufficient results), the panel will display results for each query separately.
Hide results / Show results - collapses or expands the results for a given query.
Close - hides the entire results panel.
Last updated