Workspace Isolation
Every model config, document group, document, API key, agent, and playground session belongs to the logged-in user.
Featured Project
Open-Rag-MCP
A private knowledge platform that turns user-owned documents into searchable, cited AI answers for web widgets, streaming APIs, and MCP-compatible AI clients.
RAGPrivate retrieval
MCPTool-ready access
SDKEmbeddable agents
Knowledge Workspace
Documents128
Ready94%
Agents6
Upload
Index
Retrieve
Answer
Grounded answer
Responses use group documents and can display citations for source review.
Source A0.91
Source B0.87
Group-Scoped Knowledge
Document groups separate knowledge by project, client, policy set, or product area, and integrations stay bound to one selected group.
Multi-Channel Agents
The same prepared knowledge can power an embedded web assistant, a custom streaming API client, or an MCP-compatible platform.
Product Flow
From private documents to grounded assistants
Open-Rag-MCP is organized around a setup, ingest, validate, activate, and integrate sequence.
Configure Models
Users bring their own Gemini provider keys for embedding and chat LLM configs. Embedding configs are selected when creating document groups, while chat configs can be selected for agents.
Create Document Groups
Each group acts as an isolated knowledge base with its own documents, search context, API keys, agents, and MCP access boundary.
Ingest Documents
Users add pasted text or uploaded files, track processing states, and remove documents from both the visible list and retrieval index when content becomes invalid.
Validate Retrieval
Search Bench lets users test natural-language queries, result count, reranking, diversity options, chunk text, source metadata, and relevance scores.
Publish AI Agents
Agents are attached to document groups, configured with instructions, allowed origins, history settings, citation behavior, and active status.
Architecture
One knowledge layer, three integration paths
The project is built around reusable group knowledge. Once documents are indexed and validated, developers can connect the group through the Web SDK, a server-side streaming API, or MCP.
Web SDK
Browser-safe public agent keys initialize a floating assistant widget with markdown responses and optional citations.
Streaming API
Developers can own the frontend while using private group API keys from a backend service to stream agent responses.
MCP Access
Compatible AI platforms can retrieve relevant chunks from a selected document group using authenticated MCP requests.
Agent Playground
Users test active agents, reopen saved sessions, review streamed answers, inspect citations, and validate behavior before publishing.
Trust Controls
Designed for private AI workflows
Users configure provider keys once, and raw provider keys are not shown back after saving.
Private group API keys are shown only at creation time, with later screens exposing only safe hints and metadata.
Agents, search, API keys, Web SDK access, and MCP access retrieve from the assigned document group only.
Agent settings decide whether answers show inline citation markers and expandable source details.
Functional Documentation
Read the detailed product documentation
Create an account in the Open-Rag-MCP product to explore the workflow hands-on. This section provides the detailed functional documentation for reviewing the product design and user-facing capabilities.
Detailed documentation
The documentation covers workspace access, model setup, document groups, ingestion, search validation, AI agents, integrations, playground behavior, and governance.
Project Assistant
Ask the embedded assistant about Open-Rag-MCP.
The page uses a project-specific OpenRagAgent public key, separate from the resume assistant on the homepage.