Knowledge map, LLM parameters, and RAG training
A major AI-agent platform update: a visual data graph for LLM context, compact skill settings, YML/XML product feed integration, and RAG improvements. Below — key changes worth testing in the dashboard.
Visual AI data graph and LLM context control
A visual knowledge-base map — structuring sources for the model: site crawl, RAG documents, FAQ, brief, contacts, and product catalog. Clusters by site sections, counters, and quick jumps to the right block.
On the map: drag nodes, reset layout, refresh data — easy to see what is already connected to the agent and where context gaps remain.
Site crawl for AI training and knowledge structure
Knowledge base reworked: compact section cards, settings in a side drawer. Automatic page scanning — pinned URL, extra pages, “+N new · M updated” stats in crawl history.
YML/XML product catalog for e-commerce
Direct product catalog upload as CSV, XML, and YML. The NeuronChat agent uses an up-to-date catalog in answers: prices, service attributes, and stock — relevant for online stores.
Syncing store prices with the AI agent helps consult buyers without manually updating prompts.
LLM temperature, token limit, and tone of voice
Agent settings in a new format: cards for Main, Goals & rules, Model parameters, Skills — click opens a drawer with details.
Important model parameters live in a dedicated UI: temperature (creativity), max tokens per session, context depth, and reply delay in integrations. More in Features.
Skills, lead qualification, and in-chat lead gen
Skills and Goals & rules: behavior rules save with main settings; removed an extra Apply button — less confusion when training the agent.
Fixed in-chat lead gen: leads from the app domain send correctly, [COLLECT_LEAD] markers removed from replies, phone confirmation without extra text.
RAG system and document upload optimization
Fixed RAG architecture: uploaded internal company documents and FAQ files again stably participate in agent answers.
On the knowledge map FAQ sources are separated from site pages; documents show by file name — easier navigation when configuring RAG for business.
Telegram notifications by Chat ID
Notifications section refreshed: compact Telegram card with Connected / Partial status; bot and Chat ID settings in a drawer on click.
Multi-agent without mixed context
Knowledge base, settings, skills, and widget are bound to a specific agent; switching in the Agents menu without mixing LLM context across projects.
Have an idea or a question? write to us · @jimmyneuronsupport · Documentation