Straight answers, no runaround

A platform for creating, publishing and monitoring AI assistants (chatbots) that answer from your organization’s documents and cite their sources. It covers document import, assistant and model configuration, publishing (widget, public page, email, Slack, phone, API) and administration.

Retrieval-Augmented Generation: before answering, the assistant searches your documents for the relevant passages, then writes its answer from those passages and cites them. This reduces made-up content and makes the answer verifiable.

No. Creation, configuration, document import and publishing are all done from the interface. An AI-guided creation assistant even suggests an initial configuration. The API is available for teams who want to go further.

PDF, Word (.docx), PowerPoint (.pptx), Excel (.xlsx), CSV, plain text, Markdown, HTML and JSON. With the corresponding keys: images and scanned PDFs (Mistral OCR), audio and video (transcription). Websites can be indexed with a Pro license. The legacy .doc, .xls, .ppt and OpenDocument formats are not supported.

Yes. The assistant can only cite excerpts actually retrieved from your documents; citations are numbered and shown with the document name, an excerpt and, for a web page, its URL. The embeddable widget currently displays the numbered answer without the detailed list of sources, which is provided by the chat portal, the public page and the API.

Yes, with a Pro license: the user asks for a human, an agent takes over the conversation with its full history, and the assistant then resumes. The conversation center also manages tickets, teams and response times.

Yes, with the phone callbot: the assistant answers by voice on your phone numbers, using the same knowledge as the chat, in French or English. It routes calls by topic and business hours, transfers to an agent or records a callback request. You need a SIP trunk from your carrier and keys for speech recognition and speech synthesis; call audio is never recorded.

In the infrastructure where the platform is installed: PostgreSQL database, pgvector vector database and file storage. However, questions and the necessary excerpts are sent to the chosen model provider (unless you use a self-hosted model), and embedding goes through OpenAI or Mistral.

Not entirely, at this time. Answer generation can rely on a model hosted on your own infrastructure (Ollama, vLLM), but document embedding, transcription, OCR and some internal processing use OpenAI or Mistral.

They are encrypted at rest (AES-256-GCM) in a credential vault, can be shared with named users, and are never sent to end users’ browsers. EasyRag API keys are stored only as hashes.

Retention policies archive, anonymize or delete conversations after a period of inactivity, per organization or per assistant, with a prior dry run and optional legal hold. Anonymization relies on pattern detection (emails, phone numbers, IBANs) and does not by itself guarantee the anonymity of free text.

Yes: content validation screens each document before indexing (file safety, sensitive content, personal data, prompt injection attempts). Questionable documents are quarantined until a human decides. This feature is enabled on the platform and requires a screening provider.

Yes. The platform consists of Docker containers (API, RAG service, web interface, PostgreSQL, pgvector, Redis). A license is required beyond the free tier. The Deployment page details the prerequisites and external services.

OpenAI, Mistral, Anthropic and Google Gemini, as well as Ollama, vLLM and any service compatible with the OpenAI API. The model is chosen per assistant, and the benchmark compares several models on your own questions.

No. EasyRag is an MCP client: your assistants can use tools from remote MCP servers (Pro license). The platform does not itself expose an MCP server.

There is no native integration. These tools can call the EasyRag REST API like any other HTTP service. For n8n, a workflow template is provided in the platform.

Without a license, the platform runs in a limited free mode (one knowledge base of 5 documents). A “Registered” license, free after email verification, raises the limit to 50 documents per knowledge base. The Pro license removes the limits and enables website indexing, agent tools, monitoring, human handover, the conversation center and assistant trees. Contact us for Pro license terms.

The Demo page features a demo assistant and shows how the interface works. For a trial on your own documents, contact us: we will look at your needs and the best way to meet them.

EasyRag is not a company’s proprietary software: it is a community platform, published by Patrick Beaucamp and distributed as Docker images on Docker Hub. LaTribu-IT.fr is its main sponsor. Advanced features are enabled with a Pro license.