Doks

Doks automatically discovers and indexes content from websites and documentation sites by crawling provided URLs, extracting text and structure from HTML/markdown sources, and storing normalized content in a vector database for retrieval. The system handles multi-page crawling, respects robots.txt, and deduplicates content to build a comprehensive knowledge base without manual content upload or formatting.

When a user asks the chatbot a question, Doks retrieves the most relevant content chunks from the indexed knowledge base using semantic similarity search, then passes those chunks as context to an LLM to generate a response grounded in the source material. This approach reduces hallucination by constraining the model to only synthesize information present in the training content, and includes citations or source links in responses.

Doks provides a visual interface for configuring chatbot behavior (tone, response length, fallback messages) and deploying the chatbot to websites via embedded widget, Slack, or other channels without requiring code. The system handles conversation state management, message routing, and channel-specific formatting automatically, allowing non-technical users to launch and iterate on chatbots.

Doks uses vector embeddings to convert both user queries and indexed documentation chunks into semantic representations, then ranks chunks by cosine similarity to find the most contextually relevant content for answering a question. The ranking system considers both semantic relevance and metadata (recency, source importance) to surface the best sources for LLM context.

Doks maintains conversation state across multiple turns, storing user messages and chatbot responses in a session-scoped context window. The system uses conversation history to provide coherent multi-turn interactions, allowing users to ask follow-up questions and the chatbot to maintain context without re-explaining previous answers. Context is managed per user session and automatically cleared after inactivity.

When a user question falls outside the scope of the indexed knowledge base (low confidence match or no relevant content found), Doks can be configured to provide a fallback response, suggest related topics, or escalate to a human agent. The system uses confidence thresholds to determine when to escalate rather than risk providing inaccurate information, and can route escalations to email, Slack, or ticketing systems.

Doks abstracts the underlying chatbot logic and deploys it across multiple channels (website widget, Slack bot, email integration) with channel-specific formatting and interaction patterns. The system maintains a single knowledge base and conversation engine while adapting the interface and message format for each channel, allowing users to interact with the same chatbot through their preferred medium.

Doks tracks chatbot interactions, including user questions, chatbot responses, escalations, and user satisfaction signals (thumbs up/down, ratings). The system provides dashboards showing conversation volume, common questions, escalation rates, and user satisfaction trends, enabling teams to identify gaps in documentation and optimize chatbot performance over time.

Doks tracks versions of the indexed knowledge base as documentation is updated, allowing teams to review what content changed and when. The system can re-crawl documentation sources on a schedule or on-demand, detect changes, and update the vector index incrementally without requiring full re-indexing. Teams can also manually add, edit, or remove content from the knowledge base.

Doks allows teams to configure the chatbot's tone (friendly, professional, technical) and response style (concise, detailed, with examples) through configuration parameters that are passed to the LLM as system prompts. The system applies these style preferences consistently across all responses without requiring prompt engineering or code changes, enabling non-technical users to customize chatbot personality.