EWE, the AI agent that works for your company
EWE is an autonomous artificial intelligence agent for businesses, installed on your server or on a virtual machine in Europe. It acts on your software, remembers every decision and turns every successful task into a reusable method, with the AI of your choice and always under human control.
EWE acts on your software, remembers everything and learns from every task
EWE finds any past conversation in about twenty milliseconds
Your preferences, your projects, your decisions: it remembers everything, with no history limit.
EWE turns every successful task into a reusable method
More than 200 skills from day one, and it writes new ones on its own.
EWE carries out tasks instead of explaining how to do them
Web, documents, messaging apps, business software, with the best AIs on the market, your choice.
EWE splits the work between several agents, day and night
It keeps its routines, reports back to you and waits for your green light on sensitive actions.
EWE in figures
| Indicator | Value | Detail |
|---|---|---|
| AI models available | more than 300 | OpenAI, Anthropic, Google, Mistral, DeepSeek, Qwen, Llama… or a model installed on your servers |
| Conversation channels | 25 | messaging apps, email, SMS and notifications, one shared memory |
| Built-in tools | about 100 | web, browser, documents, code, images, voice |
| Available skills | more than 200 | 58 built-in and 152 optional official ones, plus those EWE writes itself |
| Memory search | about 20 ms | full-text index (SQLite FTS5), unlimited history |
| Agents in parallel | 10 | up to 27 agents on one assignment, across 3 levels |
| Human roles (EWE Panel) | 4 | owner, admin, operator, viewer; 100% of actions logged |
| Backup (EWE Chat and EWE File) | daily | 14 archives kept |
| Technology foundation | 250,000+ GitHub stars | and 53,000+ forks as of 30 September 2026; EWE itself is proprietary software |
The EWE offer has three levels: engine, extensions, management
- The EWE engine: the heart of every agent: memory, skills, action, omnipresence, agent team and automation.
- The extensions: EWE Chat (the conversation app, with PDF, Word, Excel and PowerPoint deliverables) and EWE File (the document space shared with the agent).
- EWE Panel: the tool for running a fleet of agents: roles, budgets, human approvals, a log of every action; it also runs other agent engines.
What sets EWE apart from the AI assistants on the market
- Model-agnostic orchestrator: each task goes to the best-suited of more than 300 models, with automatic failover to another provider if one goes down.
- Closed learning loop: task, written procedure, reuse, improvement; the agent gets faster with every repetition.
- Memory split in three (episodic, semantic, procedural), stored as readable files on your infrastructure.
- Native Model Context Protocol (MCP): any compatible software plugs in with a few lines of configuration.
- Sovereign deployment: on-premise, on a European virtual machine or in a private cloud, including offline with a local model.
- Human control: sensitive actions wait for a green light, and EWE Panel logs every action and monitors every budget.
The EWE engine brings together six capabilities
Memory
It remembers you, your projects and everything you have told it.
Skills
Every successful task becomes a method it reuses.
Action
It does not explain how to do it: it does it, in your software.
Omnipresence
One single assistant, on all your messaging apps, with the same memory.
Agent team
It splits the work between several specialised agents that move forward at the same time.
Automation
Give it a routine or a goal: it takes care of it, even at night.
EWE Chat, EWE File and EWE Panel complete the engine
EWE Chat
The space where your teams work with EWE. You watch it think and work live, and the documents you asked for arrive under the answer, ready to send.
EWE File
Your agent’s shared filing cabinet. What you give it, what it produces, what the team shares: everything is organised, viewable, and usable in one sentence.
EWE Panel
The control tower for companies that scale up. Recruit agents, organise them like a team, give them a budget, and stay in charge of every decision.
With EWE, only the request leaves the company, and nothing with an AI installed on your premises
With an online assistant, every question, every file and every piece of context goes to the vendor, who decides where it is stored, for how long and with which sub-processors. EWE turns this around: it installs on your premises, and you decide what leaves, down to nothing at all.
EWE uses more than 300 models, where each assistant is limited to its own vendor’s
ChatGPT, Claude, Gemini, Copilot, Mistral, DeepSeek, Muse, Grok: each assistant confines you to the models its vendor has chosen, and often to its cloud and its apps. EWE does the opposite. It works with the best models on the market, theirs included, on your infrastructure and in your teams’ messaging apps.
EWE in technical terms: a model-agnostic, sovereign AI orchestrator
EWE is an AI orchestrator: an autonomous artificial intelligence, also known as agentic AI. It combines model-agnostic LLM routing across more than 300 models, the Model Context Protocol (MCP) to plug into any software, a closed learning loop (every successful task becomes a reusable skill) and the automation of complex workflows, on sovereign infrastructure: on-premise, a European virtual machine or a private cloud.
What EWE is not
- EWE is not open source: it is proprietary software, built on an open-source foundation.
- A 100% local AI model requires servers equipped with graphics cards.
- GDPR compliance depends on how it is used: EWE provides the means, not the guarantee.
- Pricing is on quotation (number of agents, hosting, chosen AIs).
Frequently asked questions about EWE
What is EWE?
EWE is an autonomous artificial intelligence agent for businesses, installed on your server or on a virtual machine in Europe. In practice, it acts on your software, remembers every decision and turns every successful task into a reusable method, under human control.
What is the difference between an AI agent and a chatbot?
A chatbot answers a question; an AI agent achieves a goal. In practice, EWE plans the steps, acts on your tools itself, checks the result and reports back, then remembers it for next time. This is known as agentic AI.
What are the benefits of a model-agnostic AI orchestrator?
A model-agnostic orchestrator gives each task to the best-suited model, instead of locking you into a single vendor. In practice, EWE chooses among more than 300 models, switches to another provider on its own if one goes down and adopts a new model the day it comes out, without rebuilding anything.
How does EWE learn from its tasks?
EWE runs a closed learning loop: after a successful complex task, it writes down the procedure that worked, reuses it and improves it. In practice, its memory is split in three: the conversation history (episodic), facts and preferences (semantic) and skills (procedural).
See EWE on your own use cases
A demo of what EWE would do for you, on your tools: installed on your server or in Europe, with the best AIs on the market or your own, and sensitive actions always approved by a human.