Reef is the governance platform for AI agents. Create, govern and audit every agent your company runs, wherever it was born.

Talk to Loomi See how it works

Every time zone is building agents. No dashboard shows them all.

AI has reached everyone, from the boardroom to the factory floor. Now agents are born in Copilot Studio and AI Suite across regions, with no single view of what exists, what each one reads or what it costs. Once usage is billed by the token, every duplicate agent becomes a line of cost.

New agents, over one day 00:00 UTC
Asia
Europe
Americas
How many exist?What does each one read?How much do they consume?

The agents have arrived. Control hasn’t.

Adoption is growing fast. What separates the projects that survive from the ones that get cancelled is measured cost and controlled risk.

  1. McKinseyThe state of AI, 2025

    Experimenting

    62% of organizations already experiment with AI agents, and 23% are scaling them in at least one function.

  2. GartnerAugust 2025

    Embedded

    40% of enterprise apps will embed task-specific agents by the end of 2026, up from less than 5% in 2025.

  3. GartnerJune 2025

    Cancelled

    40%+ of agentic AI projects will be cancelled by the end of 2027, over rising costs, unclear value or weak risk controls.

Prosus built 60,000 agents and found that about 2% carry the results.

In eighteen months, the 40,000 people at Prosus created more than 60,000 AI agents, and about 2% of them drive most of the business impact. Finding those agents takes data on the whole base, and Reef gives you that view from the very first agent.

60,000 agents40,000 people18 months
2 in every 100 active agents carry most of the impact. Each square is 1% of active agents (illustration). Source: Prosus, 2026.

One place to create, govern and audit every agent, wherever it’s born.

Reef runs in your own cloud. It registers the agents built in Copilot Studio and AI Suite, shows who created each one, what it reads and what it costs, and requires a human approval for every write. Agents keep running on your own models.

area tagsauthorusers per agentEnglish interface

All your agents, in one library.

Agents from Copilot Studio, AI Suite and Reef show up together, with area, author and model. Whoever needs an agent searches before building, and joins what already exists instead of creating another.

3 sources1 catalogsearch and filter by area

Illustrative interface with demo data. Try searching and filtering.

Agent library

10 agents

Governance that scales with use.

Counting users per agent is new in Reef. With it, you set the rule: restricted-use agents stay free, and agents shared by hundreds of people go through the area approver’s review.

40
Free useA restricted-use agent runs without review.

Illustrative example. The real threshold is configurable and set with each team.

Seven capabilities, working as one.

A single agent catalog

Shows what already exists, who built it and where it reads data from, before anyone builds a duplicate.

Copilot Studioagent publishedAI Suiteagent publishedReef interfaceagent created directly
Reef catalogauthor, sources and model recorded on every publish

Cost control per agent

Shows consumption by agent, area and model in real time, and locks the budget of every mission.

Accumulated cost$0.00of $20.00 budget
realbudget pacebudget lock

Human approval on writes

Blocks any write to SAP, Ariba or CLM until the area’s approver releases it.

Write request · simulation
Agent
Order orchestrator
Target
SAP · table and record shown before the decision
Impact
described in business language
Approver
the team that owns the data

Access defined by the architecture

Sets, table by table, what each agent can read, without relying on the agent itself.

ReadAgentMCP Gatewayallow / deny by scopeData Proxyread-onlySAP · Salesforce · data lakes
WriteAgentAction GatewayHuman approvalSAP · Ariba · CLM
ModelSensitive dataPresidioanonymizesYour modelAPI or self-hosted

An audit of every action

Records who called, what was read, what was written and who approved it, in an immutable log.

    The right model for each task

    Points to the cheapest model that solves each operation, and keeps legal and finance data on self-hosted models.

    Pick a task · simulation

    Recommended: light model. The team can accept or switch.

    Missions for decision makers

    Answers one-off questions about SAP data, with a set budget, without waiting in an analyst’s queue.

    Q3 margin checkIn production
    finance@company · async between time zones
    4.8ktokens used$0.42estimated cost$5.00budget
    Progress2/4 gates done

    How we’ll measure success.

    1. What isn’t in the catalog isn’t governed, so we track the share of active Copilot Studio and AI Suite agents registered in Reef every month, against the inventory built in discovery.

    2. Every repeated agent eats budget and attention. A quarterly review by area tag and description finds and merges them with your AI team.

    3. The basis for deciding what to keep, switch or switch off: tokens and spend per agent, area and model, straight from Reef’s cost dashboard.

    4. The gate has to protect without stalling the operation, so we measure the time from each request to the approver’s decision in the audit log.

    5. You can only prove what’s recorded: the share of agent calls with a complete record, checked by sampling against the gateway logs.

    6. It shows whether the catalog stops duplicate builds: users per agent, and people who join an existing agent instead of creating a new one.

    Agents will keep being born in every time zone. You can choose to see every one of them.

    Every agent, in view

    A pilot answers the question that opened this page: catalog, cost dashboard and audit trail working on real agents, inside your cloud, with the code in your hands.

    The architecture.

    OrchestrationHermes Runtime (Loomi layer) · parallel sub-agents
    BackendPython · FastAPI · PostgreSQL · Redis · per-user rate limiting
    GatewayGo · pgx
    FrontendReact · TanStack Query and Router · Radix UI · Tailwind CSS
    Identity and accessRBAC and SSO · Keycloak · your identity provider
    PrivacyPresidio Analyzer and Anonymizer (Microsoft)
    DeployDocker · Kubernetes · Helm chart

    Infrastructure

    Reef is deployed in your own cloud or on-premise, inside your account and under your security, identity and compliance policies. Loomi has no access to your data or to the agents’ outputs, and can host it for you at an additional monthly cost.

    Integrations

    Copilot Studio and AI Suite register through API, MCP or the Reef interface. SAP, Salesforce and data lakes are read through the Data Proxy, and every write goes through the Action Gateway with approval.

    Copilot StudioAI SuiteSAPSalesforceData lakesREST

    AI and models.

    1. Reef is model-agnostic and uses the subscriptions and API keys you already have, plus local models for sensitive data. The model stays outside the product.

    2. Data never leaves your perimeter. Presidio anonymizes it before anything goes to a model, in line with LGPD and GDPR, and access is defined per table and per profile.

    3. It depends on the model chosen for each area. Areas with stricter compliance, like legal and finance, use self-hosted models only.

    4. Reef’s code becomes yours, for customization and autonomy over time.

    How a rollout runs.

    Discovery and design

    Agent and user inventory with IT, integration design, a clickable prototype and a detailed roadmap.

    4 to 6 weeks

    Installation and customization

    Reef in your cloud with SSO, Copilot Studio and AI Suite agents registered, and priority integrations with SAP, Salesforce and data lakes.

    12 to 16 weeks

    Acceptance and pilot go-live

    Tests in your environment, adjustments, the pilot deploy and training for approvers and administrators.

    3 to 4 weeks

    Evolution and expansion

    New connectors, governance rules, continuous agent evaluation and expansion to other regions.

    ongoing · squad model

    Initial estimate, refined at kickoff. The lighter bars show each phase’s longest range.

    Who builds it with you.

    • EM

      Engagement manager

      Your main point of contact, owning scope, deadlines and the calendar across time zones.

    • TL

      Tech lead, Reef and integrations

      Owns the architecture of the install in your cloud and the design of every integration.

    • PE

      Platform engineer, gateways and data

      Builds the Data Proxy, the Action Gateway and the MCP Gateway over your data sources.

    • PD

      Product designer

      Designs the interface customizations, the English version and the prototype in your identity.

    • AI

      AI and evaluation specialist

      Sets up the Model Router, continuous agent evaluation and the model rules for each area.

    • DO

      DevOps / SRE

      Runs the deploy on Kubernetes with Helm, SSO and the security policies in your environment.

    • CM

      Adoption and change management

      Prepares approvers and administrators, and helps each area use the catalog before building anything new.

    We build the pilot together. Then Reef keeps evolving with you.

    After go-live, a dedicated Loomi squad stays with Reef inside your company.

    Talk to Loomi