Corthex / Feature
RAG that turns approved sources into customer-ready answers.
Corthex uses retrieval-augmented generation to find relevant source chunks before generating a response, helping assistants answer from the knowledge your team has approved.
01
Input
02
Control
03
Output
04
Measure
Page focus
Rag
RAG chatbot feature
Direct answer
RAG chatbot feature
A focused look at one operating layer inside Corthex.
Corthex uses retrieval-augmented generation to find relevant source chunks before generating a response, helping assistants answer from the knowledge your team has approved.
Written for technical and support teams evaluating retrieval quality. Search intent: Understand the RAG feature inside Corthex.
Answer from current policies and product documentation.
Reduce unsupported answers by prioritizing source context.
Inspect and refresh knowledge sources over time.
Pair retrieval with handoff when confidence is not enough.
Citable facts
Concrete signals without the noise.
Short facts make the page easier to scan for buyers, operators, and search systems.
Inputs
Documents, URLs, pasted text, and page context
Embeddings
Google text-embedding-004 with 768 dimensions
Storage
pgvector-backed source chunks
Output
Grounded answer with citations when evidence is available
Workflow
From source coverage to controlled customer answers.
The Corthex operating loop stays consistent, while the emphasis changes by page intent and audience.
Step 1
01Connect the knowledge that should ground the answer
Corthex indexes files, URLs, product context, and operational policies so the assistant has a controlled evidence layer before it ever speaks to a visitor.
- Upload documents, paste text, or import URLs.
- Chunk and embed content for vector retrieval.
- Keep source names available for answer citations.
Step 2
02Answer with retrieval, page context, and tool policy
The assistant combines the user's message with relevant knowledge, current page context, and enabled tools such as commerce lookup or live site lookup.
- Prefer exact source matches before general model knowledge.
- Route unclear or risky questions to fallback behavior.
- Show source-backed answers when evidence is available.
Step 3
03Escalate, measure, and improve the coverage loop
Human teams can take over live conversations, review unresolved questions, and turn gaps into new knowledge sources or better prompts.
- Use staff handoff for high-intent or complex conversations.
- Track questions, leads, contact history, and usage.
- Improve source coverage from real support demand.
Buyer intent
Why teams search for RAG features
Teams usually reach this page when a generic chatbot is no longer enough. They need an assistant that can answer from controlled knowledge, cite what it used, and hand off to people when confidence, compliance, or revenue risk demands it.
- Replace ungrounded answers with retrieval-backed responses.
- Make product, policy, and support content usable inside chat.
- Keep operators in control of escalation and improvement loops.
Corthex approach
Retrieval is designed as a support workflow, not a demo.
Corthex exposes source management, URL sync, chunk inspection, and citation behavior so teams can improve retrieval quality from real conversations.
- Knowledge ingestion keeps the answer surface tied to current sources.
- Page context and live lookup reduce stale responses on public sites.
- Staff console and analytics close the gap between automation and support quality.
Corthex structure
Built for clear customer answers
Each Corthex page uses a direct answer, fact table, workflow, comparison framing, related questions, and structured data. That gives buyers, operators, and AI assistants clear passages to understand and reuse.
- Short answer block near the top of the page.
- Specific feature facts instead of vague slogans.
- Internal links that make the Corthex platform map easy to follow.
FAQ
Questions this page should answer clearly.
What makes Corthex useful for RAG features?
Corthex combines retrieval-augmented generation, source citations, embeddable chat, staff handoff, analytics, and APIs so teams can deploy assistants without losing control of answer quality.
Can Corthex answer from existing knowledge sources?
Yes. Corthex can ingest documents, URLs, and pasted text, then retrieve relevant chunks when a visitor asks a question. Storefront and widget flows can also include page context.
Can a human take over when the AI should not answer?
Yes. Corthex includes staff handoff workflows so support teams can pause the bot, continue the conversation, and preserve the full context for follow-up.
Does Corthex support developer integrations?
Yes. Corthex exposes API keys, REST endpoints, streaming chat, embeddable widgets, webhook-related workflows, and connector surfaces for production integrations.