The connection to your AI
OmniAgents and OmniAgents BusinessMCP is the connection between your compatible model and OmniAgents. The conversation stays in your AI, which accesses the Core’s context and authorized tools.
INSIDE THE CORE
Nine roles coordinated on the same Core and sharing the same authorized information. Your AI is the interface; behind it, your world stays organized.
ERIC + EIGHT SPECIALISTS
Select a role to explore it. Eric brings together the specialists’ actual contributions; the routing service handles the assignment of requests.
Nine roles, not nine chats to manage.
The model can change. The Core stays.
Eric · Synthesis and verificationCollects the specialists’ actual contributions and turns them into a useful synthesis. The context remains shared.
Interactive architecture diagramYOUR AI, NOT AN IMPOSED AI
ChatGPT, Claude or other compatible models: choose the connection supported in your environment. The Core retains Knowledge, sources, memory, organization and authorizations.
Actual compatibility depends on the client and adapter. History held by an external service remains separate from the Core’s memory.
KNOWLEDGE & MEMORY
Each layer has a precise function. A product manual does not become a fact about the company; content does not grant permissions.
Shared methodology and foundational knowledge.
Versioned application guide: product sections, actions and terms.
Initial context and industry terminology.
Skills and responsibilities of specialist roles.
Private organizational facts, with origin and authorizations.
New knowledge validated from authorized sources. It does not change the model’s weights.
Chat memory · Conversation Memory
Retains messages and context from conversations managed by the Core, with links to the session and request.
Internal memory · Deep Memory
Retains consolidated context with provenance and state, so it can be retrieved in new conversations too. The environment’s data stays yours.
INTERNAL KNOWLEDGE / A GUIDE IN THE CONVERSATION
Internal Knowledge gives agents clear context about the application: sections, functions, terms and workflows. This lets the model and Eric also act as the assistant guiding you through the app.
A versioned application guide describes the environment, its functions and the available steps.
Each role uses the knowledge relevant to its task, together with information from your environment.
They explain where to start, clarify the differences between functions and guide you through the steps of your work.
“I want to connect Google Photos and use the images in my project. Where do I start?”
Application Knowledge, data from your sources and chat memory remain separate layers. The guide explains how things work; permissions determine which actions can be carried out.
TOOL LAYER
The model communicates with the Core through MCP. Connectors, APIs and web tools connect context to authorized operational capabilities. Every action goes through the required checks.
Checks that are part of the process.
Protected credentials, separate environments, permissions and action history accompany every request. You can distinguish what is complete from what requires verification.
YOUR EVERYDAY ENTRY POINT
ChatGPT, Claude or the compatible client you have configured: your request starts there and the result returns there. You don’t need to open OmniAgents to search and prepare. To confirm sending or a change, open the app on your computer.
After setup, the dashboard is where you change settings, check the environment and confirm sending and changes. The Core must be active and reachable; capabilities depend on the connection and permissions.
WORK SCENARIOS, BEYOND QUESTIONS
From context to decision: here is how you can set up a complete piece of work in the conversation with your AI.
“Read the requests that have arrived since yesterday, compare them with today’s appointments and tell me what needs attention. Prepare a reply to the two most urgent ones, but let me approve them before you send them.”
Connects senders, previous conversations, deadlines and appointments.
Separates documented urgency, routine requests and missing information.
Composes drafts using authorized context and keeps sources accessible.
“I am meeting this client tomorrow. Retrieve the latest exchanges, find the most recent proposal and compare it with the previous one. Prepare the decisions to make, promises already made and anything missing.”
Retrieves conversations, available documents and references to the meeting.
Distinguishes versions, explicit commitments, differences and unconfirmed points.
Organizes the material for decisions, with references to its origin and without filling gaps with assumptions.
“Reconstruct the purchase of my computer: order, invoice, warranty and any support requests. Tell me which documents you found, where they are and what you cannot verify.”
Uses senders, dates, order references and indexed content. OCR and other capabilities come into play when available.
Groups relevant documents, distinguishes duplicates and preserves their origins.
Orders what is documented and makes missing pieces visible.
“Prepare the invitation for Saturday’s opening with the title, date and address from the calendar, and put it in my design app.”
Date, location and images from memory and authorized sources.
Composes the draft and shows what will be created in the app.
The app is changed only after you choose “Confirm and run”.
“Follow the requests arriving on the connected channels, organize the ones to handle and prepare replies. Carry out only actions already authorized; leave the others for my confirmation. Have a summary ready for me tomorrow.”
The Business runtime handles available, configured work with the context of your business.
Agents use permitted tools and separate completed work from proposals awaiting approval.
The summary distinguishes results, receipts, tasks still queued, errors and required confirmations.
The system works on the sources you authorize. Missing information remains explicit.