Documentation
Store and organize the knowledge your bots draw on — entries, documents, and ingestion pipelines.
Knowledge bases store information your bot can reference when answering questions. When a user asks something, Neuron searches the knowledge base using hybrid search (combining semantic understanding and keyword matching) to find relevant entries and includes them in the AI's context.
Go to Knowledge and click "Create". Provide a name, description, and optionally a root instruction. The root instruction tells the AI how to use the knowledge — for example, "Use this knowledge to answer customer support questions about our products."
Entries are individual pieces of knowledge. Each entry has a title, content, and optional folder and source fields. Create entries manually for specific FAQs, product details, policies, or procedures.
Upload text documents to automatically create knowledge entries. The content is chunked into manageable pieces and stored as separate entries. This is the fastest way to import existing documentation.
Entries can be organized into folders: general (default), faqs, skills, contexts, and policies. Folders help you structure knowledge logically and make it easier to maintain.
A knowledge base can be attached to multiple bots, and a bot can have multiple knowledge bases. Go to the bot's Knowledge tab to manage attachments. When a bot has multiple knowledge bases, all of them are searched for relevant context.
Use the search feature on a knowledge base page to test how the hybrid search works. It combines semantic search (understanding meaning) with full-text search (matching keywords) for the best results.
After adding or updating entries, sync the knowledge base to regenerate vector embeddings. This ensures the semantic search is up to date. The sync runs in the background and processes all entries.
Embeddings are generated automatically for new entries, but a full sync is recommended after bulk edits or uploads.
You can ingest content into a knowledge base programmatically using the neuron_ingest_knowledge API/MCP tool. Ingestion supports optional LLM processing modes: raw (stored as-is), summarize (condenses the content), extract_facts (pulls out key facts as bullet points), or custom (your own instruction for how to process the content before storing). Use externalId for deduplication — if an entry with the same externalId already exists, it is updated instead of duplicated. Entries can be organized into folders (general, skills, contexts, documents, faqs). This is how bots self-learn from conversations and how external pipelines feed knowledge into Neuron.
When enabled on a bot, Neuron can automatically ingest knowledge from conversations. The bot analyzes conversations from configured group chats and DM phones, extracts useful information, and saves it to a designated knowledge base folder. Configure this from the bot's metadata (learningConfig) — set enabled to true, specify a target folder, and list the groupChats and/or dmPhones to learn from. Set useAdminPhones to true to learn from admin conversations automatically. This is different from Reflections — self-learning adds facts to the knowledge base directly, while reflections propose behavior changes for human review.