Knowledge Bases
Semantic search for everything, not just code
Dendrite was built for code. But teams have more than code.
Product specs. Architecture decisions. Meeting notes. Runbooks. Onboarding guides. All scattered across Google Docs, Notion, Confluence, or worse—someone's head.
Knowledge Bases brings it all together.
The problem
"Where's that doc about the payment refund process?"
Search Notion. Search Slack. Search email. Give up. Ask someone. Hope they remember.
Organizational knowledge is fragmented. Even when docs exist, finding them is painful. And keyword search fails when you don't know the exact words used.
Semantic search for documents
Knowledge Bases lets you upload any document—markdown, text, HTML—and makes it semantically searchable.
"How do we handle chargebacks?"
Even if no document contains the word "chargeback," Knowledge Bases finds the relevant sections. It understands meaning, not just keywords.
Your payment policy doc. The support runbook. The Stripe integration notes. All surfaced by semantic similarity.
How it works
Upload a document. Dendrite processes it:
- Smart chunking — Splits by headers, paragraphs, code blocks. Respects document structure.
- Local embeddings — MiniLM generates vectors on your machine. Nothing leaves your infrastructure.
- Vector indexing — pgvector stores embeddings for fast similarity search.
Search happens in milliseconds. Relevant chunks surface instantly.
RAG chat
Don't just search—ask questions.
"What's our policy on refunding annual subscriptions?"
Knowledge Bases finds the relevant chunks, builds context, and generates an answer with sources.
Not hallucinated. Grounded in your actual documents.
Click the sources to see exactly where the answer came from.
Multi-turn conversations
Knowledge Bases maintains session history.
"What about monthly subscriptions?"
It remembers you were talking about refunds. Finds the relevant policy for monthly plans. Continues the conversation.
Ask follow-up questions naturally. Build understanding incrementally.
The API
# Search
POST /api/v1/projects/:id/knowledge_bases/:slug/search
{ "query": "authentication flow", "limit": 10 }
# Chat
POST /api/v1/projects/:id/knowledge_bases/:slug/chat
{ "question": "How does SSO work?", "session_id": "..." }
Integrate Knowledge Bases into your own tools. Build custom interfaces. Power internal chatbots.
MCP tools
Claude Code gets three new tools:
kb_list — See all knowledge bases in a project.
kb_search — Semantic search across documents.
kb_ask — RAG-powered Q&A with sources.
Ask Claude: "Check the knowledge base for our API rate limiting policy."
Claude queries Dendrite. Finds the docs. Gives you the answer.
Use cases
Onboarding — New team members ask questions, get answers from your docs.
Support — Query runbooks and troubleshooting guides instantly.
Product — Search specs and requirements across projects.
Compliance — Find relevant policies without knowing exactly where they live.
Engineering — Quick access to architecture decisions and design docs.
Privacy first
Your documents stay on your infrastructure.
Embeddings are generated locally. No external API calls except Claude for chat responses.
Self-host it. Own your knowledge.
Try it
# In Dendrite dashboard
# Create a knowledge base
# Upload some documents
# Start asking questions
Finally, docs that are actually useful.
— Andres