Our vision
Enterprise AI is still rebuilt project by project: LLM orchestration, prompts, integrations, security, multi-tenancy, governance and observability. Teams spend months assembling the technical layer before delivering business value.
AgentMaurice changes that pattern. We build agentic infrastructure for companies that want to create and operate AI agents in their products and operations, under their own brand.
At the core of the system is the Agent Spec, a declarative, versioned and auditable artifact. It describes what an agent should do, with which data, tools, validations and outputs. AgentMaurice compiles that intent into executable agents, then helps teams test, deploy and reconcile the real state.
AgentMaurice is not a vertical business app replacing your product. It is the AI infrastructure layer that makes your applications agentic.
What we do
Agent Spec: Terraform for AI agents
Describe your agent declaratively: goal, inputs, tools, workflow, validations and expected outputs. The Agent Spec becomes a readable contract for product, engineering and IT teams, independent from any specific LLM model.
No disposable glue code for every experiment. A desired state, a compilation cycle, a deployment plan, then a reconciliation loop.
AgentMaurice Studio: conversational administration
AgentMaurice Studio lets teams operate the infrastructure in natural language: create a deployment, evolve an Agent Spec, diagnose an incident, understand drift or prepare a change plan. The interface keeps context and makes AI operations more accessible without removing technical control.
Native MCP Protocol
AgentMaurice is built around the Model Context Protocol. It can connect internal and external tools, RAG, structured memory, sandboxed code execution, business integrations and administration surfaces. STDIO, SSE and NATS transports make it possible to integrate existing MCP servers without locking teams into a single stack.
Governance and sovereignty
Enterprise agents must be observable, auditable and operable. AgentMaurice focuses on traceability, dedicated deployments, data control, LLM provider choice and the ability to run on the infrastructure that fits each customer.
What we sell
AgentMaurice is not sold as a shared public SaaS. Each customer keeps a clear execution perimeter, from local desktop to Kubernetes cluster.
- AgentMaurice Desktop: a local app to start, prototype and use AgentMaurice on a workstation, with a privacy-first approach.
- Managed VM instances: dedicated AgentMaurice instances operated by AgentMaurice, available in several VM sizes depending on team size, workload and expected service level.
- Enterprise Kubernetes: a dedicated deployment for regulated environments, high availability, horizontal scaling and stronger infrastructure control.
The Agent Spec remains the durable artifact while the runtime evolves: local today, VM tomorrow, Kubernetes when usage becomes critical.
The stack
AgentMaurice is built in Go for performance, reliability and deployment simplicity. The platform combines agent orchestration, multiple LLM providers, MCP, RAG, memory, OpenTelemetry observability, real-time streaming through LiveKit/WebRTC and multi-channel connectivity.
The technical choice is deliberate: provide robust, portable infrastructure for teams that need to ship to production, not only demonstrate a prototype.
Our mission
Make agentic AI as simple to integrate as well-tooled infrastructure: a company should be able to offer powerful AI capabilities to users and teams in weeks, without rebuilding the entire agentic layer internally.
AgentMaurice is proprietary software, designed and developed in France.