AgentMaurice Agent Spec describes
The spec captures the intent, resources and rules your teams can version, review and promote between environments.
Agentic infrastructure for SaaS, product, tech and IT teams
Integrate AI agents into enterprise applications in weeks, without hiring an LLM team.
AgentMaurice is the infrastructure for creating, deploying and operating product-grade AI agents under your brand, with an Agent Spec your teams can version, audit and govern.
The LLM landscape changes every quarter. With AgentMaurice, you do not ship an agent tied to one model or cloud: you compile an Agent Spec that outlives models, providers and product releases.
For personal use? Join the AgentMaurice Desktop beta.
AgentMaurice
Agent Spec It starts with a product intent.
Product
"Create an agent that reviews contracts and extracts risky clauses."
The Agent Spec is the durable artifact.
# portable · auditable · model-independent
name: assistant-contracts
model: any # gpt · claude · gemini · local
tools: [mcp://crm, mcp://docstore]
governance: { audit: full }
roles: [analyst, admin]
Agent Spec Audit Portable
Compiled, not hand-wired.
agent.bundle
Connected to your tools through MCP.
Agent
CRM
Docs
Data
Fin
Helpdesk
Web
Running in production, under control.
LIVE contract-agent ChefMaurice -> small model
Risky clauses extracted
- Non-compete clause High
- Early termination Medium
- Price indexation Medium
Every decision is traced. Rollback remains one click away.
It starts with a product intent. PM/PO teams describe the agent in natural language. No prompt-engineering project required.
2-4 weeks to ship a first production agent
x10 run-cost reduction target with model routing
100% versioned, auditable agent definitions
How AgentMaurice sees agents
Not a prompt: an executable workflow.
For AgentMaurice, an Agent combines an objective, inputs, tools, code, validations and structured outputs. Each Run is observable, auditable and can become a business mini-app when users need to intervene.
Agent executes
The Agent runs the workflow with LLM calls, MCP tools, code, sub-Agents and human approvals.
Run proves
The Run traces state, retries, costs, decisions and structured outputs so teams can operate agents like software.
AgentMaurice
Act 1 · What an Agent is An Agent is not a prompt.
Instruction text
Analyze contracts, identify risky clauses and summarize them...
Executable workflow
Defined objective
Inputs and tools
Validations
Structured outputs
An Agent is an executable workflow.
Agent structure
Objective
Inputs
Workflow
Tools
Code
Validation
Outputs
Observable Run: state · retries · traces · cost Structured outputs: answer · file · system action
Key difference: the Agent carries executable and auditable behavior, not instruction text.
The Agent mini-app : the Agent becomes an interface.
User / application
Agent mini-app
OpenUI shell
Mini-app React
the structured result feeds the AgentMaurice Run that drives the Agent
Forms, validations and dashboards connected to the Run.
Form
Dashboard
Business table
Row 1OK
Row 2Waiting
Row 3Approved
Actions · Run
Run status
running · 02:17 · 4 tools
Structured result
{ id: 124,
amount: 1250,
status: "approved" } guided formshuman validationsbusiness dashboards
An Agent is not a prompt. A prompt describes; an Agent acts through executable behavior.
Multi-team adoption
One agentic layer, four teams to convince.
The SaaS publisher remains the commercial context, but the buying committee includes product, engineering and IT leadership.
SaaS publishers
Add agentic AI to your product, under your brand, without building and staffing the whole LLM platform.
See market use casesPM / PO
Turn product intents into Agent Specs that can be reviewed, tested and iterated without waiting for a full backend cycle.
Follow the spec flowDevelopers
Keep control through MCP, APIs, CI/CD and portable specifications instead of opaque no-code builders.
See build and runDSI / CTO
Govern agents like software assets: roles, audit trail, tenant isolation, rollback and deployment choices.
Compare approaches Agent Spec
The specification that survives models, clouds and releases.
Every AgentMaurice agent is described by an Agent Spec: declarative, structured, versioned, compilable and deployable independently from the LLM provider.
capability: contract-risk-review
tenant: isolated
model: any
tools:
- mcp://crm
- mcp://docstore
governance:
audit: full
approval: human-in-loopVersioned
Specs live like product artifacts: history, diff, review, rollback and environment promotion.
Auditable
Policies, tools, model choices and runtime decisions are traced for product and security reviews.
Portable
The agent definition is not locked to one model, one cloud console or one proprietary builder.
Governable
Roles, approvals, tenant boundaries and rollback paths are part of the agent lifecycle, not an afterthought.
Build / Run
Design agents with product teams, operate them with engineering standards.
AgentMaurice separates the creation workflow from production operations so each team gets the right interface.
Build phase
Studio + MCP for creation
PM/PO can start from natural language in Studio. Developers can work from their IDE and CI/CD through MCP-compatible tools.
- No-code Studio for product discovery and iteration
- MCP-native integration for Claude Code, Cursor, Codex and internal tools
- Tests and validation attached to the Agent Spec
Run phase
ChefMaurice for production
The runtime orchestrates tools, routes calls to small models where possible and gives IT leadership the monitoring surface they need.
- Automatic model routing and cost controls
- Monitoring, audit trail and regression detection
- Deployment options for managed, dedicated or controlled environments
Product proof, not placeholder logos
The core proof is the Agent Spec itself: a concrete artifact your teams can inspect before the agent reaches production.
MCP-native by design
AgentMaurice plugs into your existing tooling surface instead of asking teams to abandon their workflow.
Ops included from day one
Monitoring, traces, model routing and rollback are treated as product requirements, not post-launch cleanup.
Why AgentMaurice
The alternative is not one tool. It is years of platform work.
Internal builds, frameworks and hyperscalers all solve part of the problem. AgentMaurice focuses on the full enterprise application lifecycle.
| Need | Internal build | Frameworks | Hyperscalers | AgentMaurice |
|---|---|---|---|---|
| Portable, versioned artifact | To build | No | Cloud locked | Agent Spec |
| No-code for PM/PO + full-code for devs | Two workstreams | Code only | No-code but closed | Studio + MCP |
| Runtime, routing and monitoring | Own platform | DIY | Cloud native | ChefMaurice + monitoring |
| Governance and audit | To operate | Mostly absent | Strong, cloud scoped | Audit, roles, rollback |
| White-label product integration | Possible | Custom build | Their brand | Under your brand |
| Deployment and data control | Depends on team | Depends on stack | Region lock-in | EU-ready paths |
The value is not another agent demo. It is the product, engineering and governance layer required to ship agents inside enterprise applications.
Use cases
Three business solutions an IT leader can scope.
Each solution starts with a concrete job, names the AgentMaurice resources to build, and connects them to the systems already used by the business.
People & organization
Keep the employee directory reliable, guide onboarding and offboarding, and answer internal questions with cited company sources.
Invoice processing
Extract invoice data, check it against supplier and purchase-order records, then route exceptions to a human before posting.
Customer Support
Classify incoming requests, draft grounded answers, and execute governed customer actions with escalation and audit.
When a solution must be reused, its Workflows, MiniApps, schemas and documentation can be packaged as a versioned Module and installed in an Application that applies roles, audit and access boundaries.
Offers
Choose the right package: Desktop, dedicated VM, or Kubernetes.
AgentMaurice is sold as a local desktop app, a managed dedicated VM sized to your usage, or an Enterprise Kubernetes deployment.
AgentMaurice Desktop
A signed desktop application for macOS, Windows and Linux, built for personal, educational and local-first workflows.
- Local runtime with embedded AgentMaurice services
- BYOK for OpenAI, Anthropic, Mistral, OpenRouter or local models
- Free non-commercial beta for makers and research
Managed VM instances
A dedicated AgentMaurice instance operated for you, available in several VM sizes according to team size and workload.
- Discover, Small, Medium and Large tiers
- Updates, backups and SLA from production tiers
- Dedicated URL, hosted credits or BYOK option
Enterprise Kubernetes
A dedicated Kubernetes deployment for regulated teams, high availability and stronger infrastructure control.
- Isolated namespace, semi-dedicated cluster or dedicated cluster
- HA, autoscaling, SSO, RBAC and audit retention
- On-premise or strict data residency possible
Next step
Pick one agentic workflow. We turn it into an Agent Spec.
The demo is most useful with one concrete workflow, one target user and one integration constraint. The primary CTA opens Calendly; email remains available if you prefer written context.