Back to vendors
R

Rubrik Agent Cloud

Also known as: Rubrik Agent Rewind, Rubrik Agent Identity, Rubrik Agent Monitor, Rubrik SAGE, Predibase, predibase.com

Visit site
Entry priceQuote only; sold through a demo requestFull pricing detail

Rubrik Agent Cloud is Rubrik's control layer for the AI agents a company runs on Copilot Studio, Bedrock, OpenAI, Gemini Enterprise, Claude and its own frameworks. It discovers agents and keeps an immutable trail from prompt to tool call, enforces plain language policies through its SAGE engine, gives each tool call a short lived token tied to a named person, blocks unscoped writes, and with Agent Rewind reverses exactly what an agent changed from Rubrik snapshots, automatically when a policy is broken.

Rubrik, the Palo Alto company listed on the NYSE as RBRK since April 2024, built its business on backup, data security and cyber recovery through Rubrik Security Cloud and now sells itself as the Security and AI Operations Company, with Adobe, The Home Depot, Pepsi and Iron Mountain among its customers. Rubrik Agent Cloud is its product for running AI agents safely, built with the team from Predibase, the model fine tuning startup Rubrik agreed to buy in June 2025, whose old address now points here.

Agent Cloud sits between applications, agents and the models behind them, and reached general availability in February 2026. Agent Monitor discovers agents across Microsoft Copilot Studio, Amazon Bedrock, OpenAI, Gemini Enterprise and Claude, plus custom agents through a standalone AI gateway and agents on endpoints, maps the data, identities and tools each one uses, flags shadow and high risk agents, and keeps an immutable audit trail from prompt to tool call.

Agent Governance enforces policies on prompts, responses and tool calls in real time; teams start from best practice templates such as PII detection or write their own rules in plain language, which the Semantic AI Governance Engine, SAGE, interprets with Rubrik's small language models and can tighten before a violation happens. Agent Identity federates with Okta and Microsoft Entra ID so each agent acts for a named person, issues a short lived token scoped to every tool call through Rubrik's MCP Gateway, and blocks writes that lack explicit policy scope before they run.

Agent Rewind reverses only what an agent changed, restoring files, databases, Salesforce records or configurations from a healthy snapshot, and SAGE can trigger a rewind on its own when it sees a violation. The Claude Code and Cowork edition adds immutable snapshots of GitHub and Azure DevOps repositories and backups of the settings that steer Claude agents, such as CLAUDE.md and system prompts. Rubrik AI, an agent built on Claude, works across Agent Cloud and Security Cloud with the customer approving each step, and Rubrik MCP opens the Security Cloud API to outside agents.

Pricing is by quote after a demo, with a self guided tour and an interactive Agent Rewind tour to explore first. Rubrik holds SOC 2 Type II for its Security Cloud platform, ISO 27001, 27017 and 27018 and ISO 42001 for Rubrik AI. The control layer runs in Rubrik's cloud platform, with the AI gateway and endpoint components placed in the path of the agents they watch. Rewinds draw on Rubrik snapshots, so recovery depends on the systems an agent touches already being protected by Rubrik.

It fits enterprises already protecting data with Rubrik that are putting Copilot Studio, Bedrock, Gemini Enterprise or Claude Code agents into production and want an undo button alongside policy and identity controls. It fits less well for teams without Rubrik backups behind the systems agents change, buyers looking for an agent builder, or anyone needing a list price.

Vendor details

Canonical URL

https://www.rubrik.com/products/rubrik-agent-cloud

Category

Security / SOC agent

Subcategory

Control layer for customers' AI agents: discovery, plain language policy enforcement, per call agent identity and Agent Rewind to undo agent changes

Funding status

Product of Rubrik, Inc., listed on the NYSE as RBRK since April 2024; built with the team from Predibase, which Rubrik agreed to acquire in June 2025.

Company status

first party product

Use cases & customers

Primary use cases

discovery and inventory of AI agents and shadow AIreal time policy enforcement on prompts, responses and tool callsper tool call agent identity with blocked unscoped writesrollback of destructive agent actions with Agent Rewindrepository and agent configuration recovery for Claude Code

Target customers

enterprise security and AI platform teams putting agents into productionexisting Rubrik Security Cloud customersengineering organizations rolling out Claude Coderegulated industries needing audit trails and recovery for agent actions

Deployment options

SaaS control layer on the Rubrik platformStandalone AI gateway in the path of custom agentsEndpoint components where agents run

Integrations

Agent Cloud covers Microsoft Copilot Studio through backend integrations, Amazon Bedrock, OpenAI, Google Gemini Enterprise, Anthropic Claude, Claude Code and Claude Cowork, custom agents on LangChain or direct model calls through a standalone AI gateway, and agents on endpoints. Identity federates with Okta and Microsoft Entra ID, recovery draws on Rubrik Security Cloud snapshots including databases, Salesforce, GitHub and Azure DevOps, and Rubrik MCP connects outside agents to the Security Cloud API.

In practice

A sales operations agent built in Copilot Studio deletes 300 Salesforce opportunities after a bad prompt. Agent Rewind ties the deletion to the agent, restores only those records from the last healthy snapshot and logs the rewind for audit.

A bank writes a plain language rule that no agent may send customer account numbers outside the company. SAGE enforces it on every response and tool call, and when a support agent's draft email includes one, the call is blocked and the rewind of its partial changes starts automatically.

An engineering team rolls out Claude Code to 500 developers. Agent Identity ties each session to the developer through Entra ID, unscoped writes to production repositories are blocked, and when an agent force pushes over a branch the repository returns to its snapshot in one click.

Agentic Index coverage score

9.5 / 14 capabilities · 68%

Integrations & Tool Calling Full

Discovery and control reach the platforms where agents are built: backend integrations for Microsoft Copilot Studio, coverage of Amazon Bedrock, OpenAI, Gemini Enterprise and Claude, a standalone AI gateway for custom LangChain or direct model builds, and endpoint deployment. Remediation acts in the systems agents touch, restoring databases, Salesforce records, files and GitHub or Azure DevOps repositories, and identity decisions draw on Okta and Microsoft Entra ID.

Sourcerubrik.com/blog/company/26/2/introducing-rubrik-agent-cloud-control-your-agents-with-airead 2026-11-18

Workflow Orchestration Full

Recovery runs as a coordinated plan: Agent Rewind correlates an agent's actions with the last healthy snapshot and restores only the affected files, data and configurations, and the Claude edition ties configuration drift to snapshots for orchestrated recovery. Rubrik AI takes an outcome the customer describes and carries it out across data, identities and agents, and Rubrik MCP lets teams save multistep recovery or compliance workflows as reusable tools.

Sourcerubrik.com/products/rubrik-agent-cloudread 2026-11-18

Knowledge Grounding & RAG Partial

SAGE judges each interaction against the customer's own plain language policies and the map of data, identities and tools behind every agent, and Rubrik AI reasons over the customer's protected data, identities and deployed agents. Agent Cloud does not give the governed agents a retrieval store of company documents, and Annapurna, Rubrik's service for feeding protected data to AI apps, is open only to qualified partners.

Sourcerubrik.com/products/rubrik-agent-cloudread 2026-11-18

Human Oversight & Guardrails Full

Guardrails act before harm: policies apply to prompts, responses and tool calls in real time, writes without explicit scope are blocked and the denial is logged against the user and agent, and SAGE suggests tighter wording for ambiguous rules before a violation occurs. When something gets through, Agent Rewind undoes only that agent's changes, and Rubrik AI asks the customer to approve each step it takes.

Sourcerubrik.com/blog/technology/26/8/introducing-rubrik-agent-identity-identity-for-agents-control-for-actionsread 2026-11-18

Security, Identity & Governance Full

Agent Identity federates with Okta and Microsoft Entra ID so every agent acts on behalf of a named person rather than a shared service account, and each MCP tool call passes three checks: SAGE on intent and parameters, the MCP Gateway on access policy, and a short lived token scoped to that single call.

Rubrik MCP keeps parity with Rubrik's role based access controls and aligns its safeguards with the OWASP MCP Top 10. Rubrik holds SOC 2 Type II for the Rubrik Security Cloud SaaS platform, ISO 27001, 27017 and 27018, ISO 42001 for Rubrik AI and CSA STAR Level 2, and its government cloud carries FedRAMP Moderate.

Sourcerubrik.com/blog/technology/26/8/introducing-rubrik-agent-identity-identity-for-agents-control-for-actionsread 2026-11-18

Observability & Auditability Full

An immutable audit trail runs from prompt to tool call, enriched with data, identity and application context, and the inventory shows when each agent was created, how much it is used, its risk profile and the tools and data it reaches. Blocked writes are logged and attributed to the user and agent, every rewind is recorded for audit, and shadow and high risk agents are flagged as they surface.

Sourcerubrik.com/products/rubrik-agent-cloudread 2026-11-18

Memory & State Persistence Partial

Durable state here is protective: immutable snapshots of the systems agents change, an immutable audit trail of every prompt and tool call, and in the Claude edition versioned backups of the files and settings that steer agents, such as CLAUDE.md, system prompts and tool access. The governed agents get no memory store from Agent Cloud, and continuity of their behavior stays with the platforms that run them.

Sourcerubrik.com/company/newsroom/press-releases/26/rubrik-launches-rubrik-agent-cloud-for-anthropics-claude-coderead 2026-11-18

Deployment & Data Residency Partial

The control layer runs in Rubrik's cloud platform as the Agent Cloud data plane, while the standalone AI gateway sits in the traffic path of custom agents and endpoint components install on the devices where agents run. Rewinds draw on Rubrik snapshots that customers already keep on premises or in cloud under Rubrik Security Cloud, and telemetry from every agent flows into the cloud layer for policy decisions and audit.

Sourcerubrik.com/blog/company/26/2/introducing-rubrik-agent-cloud-control-your-agents-with-airead 2026-11-18

Prebuilt Agents / Templates / Packs Full

Teams start from best practice policy templates and predefined guardrails such as PII detection rather than writing every rule, and Rubrik ships packaged editions for specific agent platforms: Agent Cloud for Claude Code and Claude Cowork, and an edition for Gemini Enterprise. Rubrik AI arrives as a ready agent across Agent Cloud and Security Cloud with no build step.

Sourcerubrik.com/products/rubrik-agent-cloudread 2026-11-18

Triggers & Channel Coverage Full

Enforcement and recovery start without a person: SAGE evaluates every prompt, response and tool call as it happens, write operations without policy scope are stopped before they execute, and a rewind can fire automatically when SAGE detects a violation. Discovery scans the environment continuously, so new and shadow agents join the inventory as they appear and pick up the policies that apply to them.

Sourcerubrik.com/products/rubrik-agent-cloudread 2026-11-18

Model Flexibility & Routing Partial

Rubrik picks the models behind its own work by job: small language models it runs enforce custom policies, and Rubrik AI is built on Anthropic's Claude. Customers cannot choose or swap either, while the governed agents keep whatever models they already use through OpenAI, Bedrock, Gemini or Claude, which Agent Cloud watches and controls rather than routes between.

Sourcerubrik.com/blog/company/26/2/introducing-rubrik-agent-cloud-control-your-agents-with-airead 2026-11-18

APIs / SDKs / MCP Extensibility Partial

Rubrik MCP, generally available since October 2026, reads the Rubrik Security Cloud schema in real time and exposes the full API to outside agents, with read access to protection, anomaly and compliance data plus a curated set of confirmed actions, permissions that mirror Rubrik's access roles and a service account for sign in. Its tools follow the live schema rather than a fixed list, and Agent Cloud's MCP Gateway governs customers' own MCP calls.

Sourcerubrik.com/blog/technology/26/10/introducing-rubrik-mcp-bring-your-own-agent-to-your-cyber-resilience-dataread 2026-11-18

Testing, Debugging & Optimization Not documented

SAGE can suggest refinements to ambiguous guardrails before a violation occurs, which tunes the rules rather than testing the agents. There are no test sets, red team runs or scored evaluations of the customer's agents, and agent quality shows up in risk profiles and policy violations once they run in production rather than before they ship to users.

Sourcerubrik.com/products/rubrik-agent-cloudread 2026-11-18

Browser / Computer-use Not documented

Agent Cloud discovers and governs agents on endpoints but drives no browser or desktop itself, and Rubrik AI works through Rubrik's own platform and APIs. Agents that operate screens are watched through their telemetry and tool calls rather than replicated by Rubrik, so no Rubrik agent clicks through an application on a user's behalf. Screen work stays with the agents being governed.

Sourcerubrik.com/products/rubrik-agent-cloudread 2026-11-18

The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded

Pricing

Quote only; sold through a demo request

Enterprise subscription quoted per deployment

What is public

Rubrik Agent Cloud is sold by quote after a demo request. A self guided tour and an interactive Agent Rewind tour are open without a sales call.

Billing mechanics

Agent Cloud is quoted as an enterprise subscription, with separate editions for Claude Code and Cowork and for Gemini Enterprise.

Cost watchouts

Agent Rewind restores from Rubrik snapshots, so undoing an agent's changes to a database, SaaS app or repository needs those systems protected by Rubrik Security Cloud as well.

Variable cost rationale

Contracts are quoted subscriptions rather than metered agent usage.

Additional watchouts

No list prices appear for Agent Cloud, its modules or its platform editions.

Sales call required

Yes, required for paid access

Free / trial

No free plan or trial; self guided tour and interactive Agent Rewind tour

Lowest paid plan

None; Agent Cloud and its editions are quoted

Commercial notes

Global systems integrators including Cognizant, Deloitte, LTM, HCLTech, NTT Data and Wipro deliver the Claude Code edition to their clients.

Missing data

Rubrik gives no prices for Agent Cloud, its modules, the platform editions or the Rubrik Security Cloud protection that rewinds rely on.

Agentic Index verified 2026-11-18

Alternatives to Rubrik Agent Cloud

Other products that do the work Rubrik Agent Cloud does, and what each one is.

Contact us

Found a vendor we missed? Have feedback on the index? We'd love to hear from you.