Platform and cloud
Open source at the core, deployed in the cloud you already trust.
Etienne, our open-source agent harness
Etienne wraps coding-agent SDKs such as the Claude Agent SDK with what business use needs. It is MIT-licensed, so you can read every line.
View on GitHub
What it adds to the agent loop
- Agent skills with a company skills store
- Project-scoped knowledge and decision graphs
- MCP servers per tool group, MCP UI and A2A
- Artifact viewers and editors beside the chat
- Budget tracking with limits
- Scheduled and recurring tasks
- Role-based access: admin, user, guest
- Checkpoints and OpenTelemetry traces
Your Azure or AWS account
We deploy into your own subscription, in the region you choose, behind your identity provider. Your data and your experts' knowledge stay under your control.
Microsoft Azure
- Azure App Service and Foundry hosted agents
- Microsoft Entra ID for sign-in
- Azure Key Vault for secrets
- OneDrive and SharePoint as data sources
Amazon Web Services
- Hosting in your AWS account
- Amazon Cognito for sign-in
- Amazon Bedrock Guardrails for input and output checks
Where data must stay on site, Etienne can route requests to local models.
Rent the frontier model, own the control around it
A language model is the easy part to rent. Identity, permissions, budgets, isolation and logging around long-running agents are operating-system questions. Those you should be able to read, run and explain yourself.
Agents never hold the keys
An agent talks to a gateway, and the gateway decides which tools it may use. If the agent is tricked by a malicious text or makes a mistake, the damage stays small.
Approval without watching a screen
When a decision is due, the agent sends an e-mail with a link to it, keeps its state while it waits and continues when you answer. A rejected change is rolled back.
A simple test for your team
- Could we swap the model provider next quarter without rebuilding?
- Could we explain, step by step, what an agent did last week and who approved it?
- If an agent were manipulated tomorrow, how much of our data could it reach, and how fast could we stop it?
Repositories
ears-sysmlv2-assistant
MITExample of specification-driven AI design based on international standards in the automotive and manufacturing industries.
View on GitHubwrigglets
Experimental operating system for agents on Docker Desktop or Azure: coordination through A2A, durable workflows and human approval.
View on GitHubArticles on LinkedIn
How to build an agentic OS while keeping the frontier within your reach
Read the ArticleFive AI Agents, One Sofa Factory: A Glimpse into the Agentic Enterprise
Read the ArticleUnderstanding Etienne: Complementing Claude Agent SDK's Agentic Loop
Read the ArticleFive Ways to Render Things in Agentic User Interfaces - and How to Pick
Read the ArticleFearing Etienne: How Forward-Deployed GenAI Engineers Will Replace Consulting and Software Development Teams
Read the Article
