Cody

Cody implements a retrieval-augmented generation (RAG) pipeline that accepts user queries, searches an indexed knowledge base of uploaded documents and crawled websites, retrieves the top 10 most relevant documents using semantic similarity, and generates contextual answers with inline source citations. The system maintains conversation history to provide context-aware responses across multiple turns within a session, enabling follow-up questions and clarifications without re-specifying domain context.

Cody supports three knowledge base input methods: direct document upload (PDFs, text files), automated website crawling (recurring crawls of specified domains), and API-based content ingestion. The system indexes uploaded content and crawled pages into a searchable knowledge base, with tier-dependent limits on document count and website crawl depth. Website crawling can be configured to run on a recurring schedule, enabling knowledge bases to stay synchronized with updated documentation.

Cody supports brainstorming and ideation workflows by maintaining conversation context across multiple turns, enabling users to iteratively refine ideas and explore variations. The system can generate multiple options, provide feedback on ideas, and suggest improvements based on organizational context from the knowledge base. Users can ask follow-up questions, request alternatives, or pivot to new directions without losing context.

Cody can assist with technical troubleshooting by searching support documentation, knowledge base articles, and FAQs to provide step-by-step solutions to common problems. The system retrieves relevant troubleshooting guides and error documentation, synthesizes solutions, and provides source citations so users can verify and follow detailed instructions. This capability is particularly useful for support teams handling repetitive technical issues.

Cody abstracts multiple underlying language models (GPT-4 Mini, GPT-4, Claude 3.5 Sonnet) behind a unified interface, allowing users to select which model powers their queries. Each model consumes a different number of credits per query (GPT-4 Mini: 1 credit, GPT-4: 10 credits, Claude: unspecified), with monthly credit allowances varying by tier (Basic: 2,500/month, Premium: 10,000/month, Advanced: 25,000/month). Users can switch models per-query or set a default, enabling cost-performance tradeoffs without changing application code.

Cody can be deployed as an embeddable web widget on external websites, shared via direct links, or displayed as a popup modal. The widget maintains the same knowledge base and conversation context as the web interface, enabling organizations to expose their AI assistant to customers, employees, or partners without requiring them to visit a separate domain. Widget configuration (appearance, positioning, behavior) is managed through the Cody dashboard.

Cody includes pre-built workflow templates optimized for HR functions such as employee onboarding, candidate screening, and policy question answering. These templates provide standardized prompts, knowledge base structures, and conversation flows that reduce setup time and ensure consistent responses across HR processes. Templates can be customized with company-specific policies, job descriptions, and evaluation criteria.

Cody maintains conversation history within a session, enabling the assistant to reference previous messages and provide context-aware responses to follow-up questions. Conversation logs are retained for 14-90 days depending on tier (Basic: 14 days, Premium: 30 days, Advanced: 90 days), allowing users to review past interactions. However, context does not carry across separate conversations or sessions; each new conversation starts with no memory of previous interactions.

Cody exposes an API for programmatic access to chatbot functionality, available on all subscription tiers. The API enables external applications to submit queries, retrieve responses, and manage knowledge bases without using the web interface. However, API documentation is not publicly available; endpoint specifications, authentication methods, rate limits, and request/response schemas are undocumented, requiring direct contact with Cody support for integration details.

Cody claims to support queries and responses in multiple languages, enabling non-English speakers to interact with the knowledge base. The system automatically detects query language and generates responses in the same language. However, the specific languages supported, translation quality, and multilingual knowledge base handling are not documented.

Cody supports team collaboration by allowing multiple users to access and interact with the same knowledge base and chatbot instance. Access is managed through team member invitations, with tier-dependent limits on the number of team members (Basic: 3, Premium: 10, Advanced: 30). However, role-based access control (RBAC), permission granularity, and audit logging for team actions are not documented.

Cody can generate creative content such as emails, marketing materials, social media posts, and other written content based on user prompts and knowledge base context. The system combines knowledge base retrieval with generative capabilities to produce contextually relevant content that aligns with organizational voice and policies. Generated content can be edited and refined through iterative prompts within a conversation.