# AAES (Autonomous Agentic Enterprise Systems) Operator: AAES LLC, a Delaware limited liability company. Registered address for legal notices: 8 The Green STE B, Dover, DE 19901, United States. Contact: support@aaes.ai. > The operating and governance layer for your AI workforce > > Let AI agents work. Keep your staff in control. AAES builds AAES Operate and > AAES Govern, sold separately or together as AAES Operate + Govern. Both run on the client's infrastructure or are hosted by AAES. AAES > Operate creates and runs agents, assigns private tasks and keeps run > history. It has private-work permissions, operational reviews and recorded-spend limits. It does not include > AAES Govern's controls. AAES Govern checks and records the agent actions routed > through it: permissions, approvals by authorized persons, and spending limits, > with offline record integrity checks. AAES Infer gives agents model access > with jurisdiction controls through the client's own Google Vertex AI or Azure > account. ## The problem Agent work needs defined tasks, accountable owners and controlled access. AAES Operate brings agents into your organizational structure and gives their work projects, tasks, skills and connectors. AAES Govern adds ownership, access, guardrails and an audit trail for the actions routed through it. ## Products - [Product facts](https://aaes.ai/about.html#products): Product purposes, availability and deployment options, with links to scope and execution-path evidence. The same summaries are available as [JSON](https://aaes.ai/facts.json). - AAES Infer: https://aaes.ai/products/infer.html Qualification required. Model access with jurisdiction controls and contractual limits on customer data use. Infer runs through your own Google Vertex AI or Azure account, under your own provider exemption. Your provider bills your organization directly. The API offers text chat, native Messages and native Responses interfaces, with streaming and client tools where the signed offer supports them. API scope: https://aaes.ai/developers/infer.html. Each model and processing jurisdiction needs its own qualified offer. US and EU offers work at jurisdiction level and can use several regions or member states; the other seven markets are single-country or single-jurisdiction offers. A route qualifies only with a regional endpoint inside the jurisdiction, or a Google Vertex AI us or eu multi-region endpoint on a jurisdiction-level US or EU offer. Every route also requires signed reviewer approval of no training, no secondary use, no human review and zero retention, a current signed provider observation and a published qualification. Providers are chosen per model and jurisdiction. Operate and Govern model-source support does not establish Infer availability. - AAES Operate: https://aaes.ai/products/operate.html Software for creating and running agents, self-hosted or hosted by AAES, with an org chart, roles and reporting lines, and agents using your own model accounts, projects and a shared task board, a versioned skills library, connectors and MCP servers that agents use directly with per-agent permissions, operator-enabled plugins, scheduled jobs, budgets with automatic stops on recorded spend, approval and review steps, agent configuration and routine revisions you can roll back, and installable team templates. AAES supplies a reference deployment configuration and upgrades, with assisted installation and optional enhanced support. Operate does not include AAES Govern. Govern adds spending reserved before authorization, signed decision records you can check offline, and governance for agents on other platforms. Operate is built on open-source components; the component inventory and license notices are provided in the evaluation package. AAES Operate and AAES Govern use your organization’s model-provider accounts. In standalone Operate, an agent calls your model account directly when both use the same API, such as Claude Code with an Anthropic Messages endpoint or Codex with an OpenAI Responses endpoint. Other sources are available through AAES Govern. Connect agents you already run over HTTP to give them work. A new agent is created in Operate, approved where your rules require it, then given its first task. Agent workspaces: Microsoft 365 (works across Microsoft 365 apps), Google Workspace (works across Google Workspace apps), Meetings (joins Teams, Zoom and Meet calls), a mailbox (its own email address), chat (a Teams or Google Chat handle), a phone number (its own phone number), a browser (an isolated web browser), a desktop (a virtual desktop for legacy apps) and a payment card (a virtual payment card). In standalone Operate, agents can use Browserbase browser sessions and their own Microsoft 365 or Google Workspace mail, calendar and selected files. Sending mail and invitations, sharing and changing files, cards, and outbound calls and SMS to the public need AAES Govern. - AAES Govern: https://aaes.ai/products/govern.html Connect an existing agent to Operate to give it work, and to Govern to control its actions and model calls. The agent keeps running where it runs today and can route its model calls through Govern with its own gateway credential. Govern checks permissions, required approvals and spending limits on routed actions, and applies its controls only to actions routed through it. Agents on Amazon Bedrock Agents, Bedrock AgentCore, Vertex AI Agent Engine and Salesforce Agentforce can also be registered. Govern covers their tool calls only, because their model calls stay with the vendor. With AAES Operate, each employee who directs agents can have a personal AI budget. Outcome checks read the target back after an action runs and seal the result as confirmed, mismatch or unknown. Value reports show workflow measurements against a baseline your organization sets. Its model gateway accepts the Anthropic and OpenAI APIs and routes each call by model name. Govern’s model gateway reaches model vendors, cloud platforms, gateways and self-hosted servers. The Models page lists every source and how it was tested. For any agent that sends its model calls through it, Govern’s model gateway applies model policy, holds your provider keys, checks spend before each call and seals a record after it. It translates Anthropic Messages and OpenAI Responses calls for sources that accept only chat completions. Codex agents that use Govern’s MCP tools need a source that accepts OpenAI Responses calls. Live calls through the gateway have passed for 8 sources, some only for certain call formats; the rest are tested against recorded responses or not verified by AAES. The Models page shows each result. Test each source with your own agents before you rely on it. ## Integration guides - Operate + Govern integration: https://aaes.ai/products/operate-govern.html Use both products together. Operate runs the work; Govern checks and records the model and tool calls that Operate's agents route through it. Govern supplies the agents' tool connections, and Operate turns off any tool connection that does not lead to Govern. A remote MCP server, HTTP app, plugin tool or skill added in Operate becomes a Govern draft, and agents use it only after it is reviewed and approved. Approval requests from Govern appear in the manager's Operate inbox. In AAES Operate + Govern, every model call from a Claude or Codex agent that Operate runs goes through Govern. In AAES Operate + Govern, workspace administrators can read each agent’s governance status as JSON. Fully governed: its tool calls go through Govern, its workspace’s block on coordinator plugin tools is in force with no other tool policy able to come before it, and Govern has decided a model call made with its gateway credential in the last day (the window is configurable). Actions governed: its tool calls go through Govern, but that plugin-tool block is missing or another tool policy may come before it (so plugin tool calls may run outside Govern), it has no gateway credential, or Govern hasn’t decided a model call from it in that window. Not governed: AAES Operate + Govern doesn’t admit the agent’s type and pauses it, or Govern hasn’t issued its credential yet. Only standalone Operate admits Cursor, GitHub Copilot and Meta’s Muse agent. They send their model calls to their own vendors, so Govern doesn’t govern them. ## What AAES Govern does - Registers agent identities with at least one accountable manager, ultimately answerable to a person. An approver is the person authorized to decide a particular request; the roles are distinct. - Checks actions routed through AAES Govern against a deterministic policy bundle. - Requires approval by an authorized person for actions classified as irreversible. Policies can require approval for other actions too. An agent cannot approve itself. - For supported spending requests routed through it, reserves the evaluated amount before authorization and refuses any request that would exceed the remaining ceiling. This limits what Govern authorizes. It does not cap final provider charges or spending outside AAES Govern. - Uses an amount only with evidence from the capability's registration: a per-action cost ceiling, or a request contract that binds the amount to the outgoing payload. With neither, a stated amount is refused, a capability that declares pay is refused, and an action with no amount arms every money threshold on its rule. - Any administrator can engage the emergency stop at once. Lifting it, offboarding an agent and publishing a rule each need a fresh second factor. Offboarding an agent is final. - Records whether each request was allowed, refused or held for approval. Each sealed decision can be opened and read in Govern. - Produces sealed decision and outcome records that can be exported and checked for integrity offline, without connecting to an AAES service. AAES Operate alone does not produce these records. ## Enforcement and evidence boundaries When an agent retains its own direct credentials, Govern can record reported activity but cannot prevent the call. Observation-only records do not establish enforcement. Offline integrity verification is distinct from independent attestation. Independent attestation requires a configured external witness or timestamp authority and trust material obtained separately from the export. Neither is connected in the default deployment, and the customer_hosted profile refuses to start without one. A valid record does not prove that work outside AAES Govern was captured or that a dispatched call achieved its intended result. If AAES Govern cannot write its decision journal, new requests are denied and no new grant is issued. A journal failure does not itself revoke an existing grant; grants expire within fifteen minutes unless revoked earlier. If you run the deployment, you operate its availability; if AAES hosts it, the agreement sets those responsibilities. The service level agreement commits to 99.9% monthly uptime for deployments hosted by AAES. ## AAES Govern interfaces and deployment - HTTP API: OpenAPI 3.1, authenticated by tokens issued by the deployment operator. - MCP: connect through aaesmcp (MCP adapter) using stdio or POST-only HTTP. - SDKs: the Python package aaes-govern-sdk 0.1.3 is published on PyPI, and the JavaScript/TypeScript package @aaes-ai/govern-sdk 0.2.0 is published on npm. Earlier releases remain under aaes-sdk and @aaes-ai/sdk. Other languages call the HTTP API using the OpenAPI contract. Calling the SDKs requires a configured deployment and operator-issued token. - CLI: aaesverify v0.3.0 (public Apache-2.0 offline export verifier); aaesctl (operator CLI, including the records-sidecar binding check); aaescp (local planner). - Browser console: approvals, agent identities, policies, budgets, and records. Appearance is stored per tenant, not per user. - Deployment: The supported installs are Docker Compose on one Linux host and Kubernetes, tested on Google Kubernetes Engine and in CI. Installation is assisted. It can run in an isolated on-premises environment. AAES-hosted production runs across zones for high availability. Customer-hosted high availability requires a cluster that can power machines off through an API. Both deployment types use backup and restore. aaes-setup recovery backs up and restores, and aaes-setup rollback returns a Compose install to its previous release. A restored install starts no writer until it is activated. - Production install: the customer_hosted profile requires HashiCorp Vault for provider keys, an external witness or RFC 3161 timestamp authority, and joint acceptance signed offline by the AAES owner and a person the customer names. - Sign-in: people sign in at their identity provider directly, through OIDC or SAML, including Microsoft Entra. WorkOS sign-in is refused because it cannot complete step-up; WorkOS Directory Sync stays available. Directory sync maps groups to workspaces by their fixed IDs, so renaming a group changes no one's access. - Private addresses: remote connections dial public addresses only unless an operator names the host in AAES_REMOTE_HTTP_PRIVATE_HOSTS. Operate and Govern can also be hosted by AAES. No SOC 2 report or penetration-test report exists. AAES has a documented scope and test plan for an independent penetration test. AAES Operate is self-hosted or hosted by AAES. Its setup documentation is provided in the evaluation package. ## Integrations Live tenant validation is part of a paid evaluation. Model-gateway live results are on https://aaes.ai/models.html. Request the coverage matrix for the workflows and vendors being evaluated. ## Status and access Operate and Govern are priced in headcount bands, from under 500 employees to 300,000, and we quote larger organizations individually. Users and agents are unlimited. AI model charges are separate. Start with a free workflow assessment. A paid evaluation is then scoped and priced for your workflow. Production pricing is shared on request. The customer-run platform evaluation package is included only in a paid, scoped evaluation. Operate and Govern are sold separately or together. Installation is assisted. Training and enhanced support are optional and agreed separately. Public documentation, the Apache-2.0 verifier/sample, and published Python/TypeScript SDKs are available separately and do not grant a platform license or running deployment. Evaluations run in the customer's environment, or AAES hosts them for a fixed fee. Production deployments can be hosted by AAES. Request the platform evaluation package or workflow support by emailing hello@aaes.ai or using the contact form. Public documentation describes the interfaces; it does not grant access to a running deployment, a support agreement, or the private repository. ## Links - Homepage: https://aaes.ai/ - AAES Infer: https://aaes.ai/products/infer.html - AAES Operate: https://aaes.ai/products/operate.html - AAES Govern: https://aaes.ai/products/govern.html - AAES Operate + Govern: https://aaes.ai/products/operate-govern.html Agent workspaces: Operate gives each AI employee structure and tools. Govern controls what it does with them. The tools are Microsoft 365, Google Workspace, Meetings, a mailbox, chat, a phone number, a browser, a desktop and a payment card. In standalone Operate, agents can use Browserbase browser sessions and their own Microsoft 365 or Google Workspace mail, calendar and selected files. Sending mail and invitations, sharing and changing files, cards, and outbound calls and SMS to the public need AAES Govern. Tools are created in your organization's own accounts. - About AAES and publishing standards: https://aaes.ai/about.html - Developer docs: https://aaes.ai/developers.html - Setup options: https://aaes.ai/developers/setup.html - Infer API scope: https://aaes.ai/developers/infer.html - Infer OpenAPI: https://aaes.ai/developers/infer/openapi.yaml - Operate documentation: https://aaes.ai/developers/operate.html - Govern documentation: https://aaes.ai/developers/govern.html - Operate + Govern integration: https://aaes.ai/developers/operate-govern.html - Agent platform comparisons: https://aaes.ai/compare/ai-agent-platforms.html - Agent governance comparison: https://aaes.ai/compare/ai-agent-governance.html - Native control planes compared: https://aaes.ai/compare/native-control-planes.html - Models: https://aaes.ai/models.html AAES Operate and AAES Govern use your organization’s model-provider accounts. The page lists all 29 model sources Govern's gateway reaches, grouped as vendors, cloud platforms, gateways and self-hosted servers, with each source's test evidence, and the agents that use their own models. - API: https://aaes.ai/developers/govern/api.html - MCP: https://aaes.ai/developers/govern/mcp.html - CLI: https://aaes.ai/developers/govern/cli.html - SDKs: https://aaes.ai/developers/govern/sdks.html - SDK methods: https://aaes.ai/developers/govern/sdk-contract.html - White label: https://aaes.ai/developers/govern/white-label.html - Evaluation: https://aaes.ai/developers/evaluation.html - OpenAPI: https://aaes.ai/developers/govern/openapi.yaml - Trust center: https://aaes.ai/trust.html - Security: https://aaes.ai/legal/security.html - Vulnerability disclosure policy: https://aaes.ai/legal/disclosure.html - Privacy: https://aaes.ai/legal/privacy.html - Subprocessors: https://aaes.ai/legal/subprocessors.html - Website terms: https://aaes.ai/legal/terms.html - Contact: https://aaes.ai/contact.html - Free workflow assessment: https://aaes.ai/contact.html. A 30-minute call about one workflow you choose; within 2 business days, a one-page written assessment of which steps an AI agent could take, which need a person's approval, which controls apply and what a paid evaluation would cover. - Book a product demo: https://aaes.ai/contact.html - Capability library, the index of every package page: https://aaes.ai/capability-library.html - Public organization profile: https://github.com/aaes-ai ## Library - [Evidence library](https://aaes.ai/library.html): Public product evidence, remaining gaps, and an AIUC-1 note. AAES Govern can be client-operated or hosted by AAES. It records routed action decisions and exports evidence for offline integrity checks. It is not an AIUC-1 certification. - [Capability library](https://aaes.ai/capability-library.html): Pre-built registration templates for selected vendor capabilities: 95 reviewed application packages and 13 in review. Detail pages list actions, setup requirements, risk tiers, and review sources. Install through reviewed console batches or CLI change control; test live connectivity separately. - [Scope and status](https://aaes.ai/library/scope.html): Product boundary, enforcement limits, and whether a workflow can be evaluated. - [Action controls](https://aaes.ai/library/controls.html): Named approval, capabilities, grants, budgets, fail-closed behavior, and sealed records, with gaps. - [Offline verification](https://aaes.ai/library/verification.html): How to verify an export and what a valid record does not establish. - [Control evaluation](https://aaes.ai/library/evaluation.html): Public evaluation procedure. Client results are produced during a scoped evaluation; the public verification demonstration shows record integrity checks, not customer acceptance results. - [AIUC-1](https://aaes.ai/library/frameworks/aiuc-1.html): Seven selected AAES Govern record properties. Not a certification and not an assessment of AIUC-1's requirement catalog. - [NIST AI RMF context note](https://aaes.ai/library/frameworks/nist-ai-rmf.html): Narrow note on action-layer contribution. Not a law and not an AI RMF implementation. - [CSA AARM note](https://aaes.ai/library/frameworks/csa-aarm.html): The CSA working draft AAES builds against, with AAES's dated position and the organizational rows still open. Readiness against a pinned draft, not a passed assessment. - [Jurisdiction context notes](https://aaes.ai/library.html#jurisdiction-context): Dated notes on AI-governance instruments in Hong Kong, Singapore, Australia, the European Union, Japan, Switzerland, the United Arab Emirates, the United Kingdom, and the United States. Not legal advice. ## More pages - Candidate workflows by industry: https://aaes.ai/industries.html - Export and evidence specification: https://aaes.ai/spec.html - Operate API: https://aaes.ai/developers/operate/api.html - Operate CLI: https://aaes.ai/developers/operate/cli.html - Operate MCP: https://aaes.ai/developers/operate/mcp.html - Operate SDKs: https://aaes.ai/developers/operate/sdks.html - Operate white label: https://aaes.ai/developers/operate/white-label.html - Trust model: https://aaes.ai/library/trust-model.html - Capability matrix: https://aaes.ai/library/capability-matrix.html - Evidence field map: https://aaes.ai/library/evidence-field-map.html - Framework evidence downloads: https://aaes.ai/library/framework-evaluation.html - CoSAI agentic security note: https://aaes.ai/library/frameworks/cosai-agentic-security.html - IETF WIMSE and AIMS note: https://aaes.ai/library/frameworks/ietf-wimse-aims.html - ISO/IEC 42001 note: https://aaes.ai/library/frameworks/iso-iec-42001.html - MITRE ATLAS note: https://aaes.ai/library/frameworks/mitre-atlas.html - NIST CAISSI and NCCoE note: https://aaes.ai/library/frameworks/nist-caisi-nccoe.html - OWASP agentic note: https://aaes.ai/library/frameworks/owasp-genai-agentic.html - OWASP LLM Top 10 note: https://aaes.ai/library/frameworks/owasp-llm-top10.html - Australia note: https://aaes.ai/library/jurisdictions/australia.html - European Union note: https://aaes.ai/library/jurisdictions/european-union.html - Hong Kong note: https://aaes.ai/library/jurisdictions/hong-kong.html - Japan note: https://aaes.ai/library/jurisdictions/japan.html - Singapore note: https://aaes.ai/library/jurisdictions/singapore.html - Switzerland note: https://aaes.ai/library/jurisdictions/switzerland.html - United Arab Emirates note: https://aaes.ai/library/jurisdictions/united-arab-emirates.html - United Kingdom note: https://aaes.ai/library/jurisdictions/united-kingdom.html - United States note: https://aaes.ai/library/jurisdictions/united-states.html ## Solutions - [AI governance platform](https://aaes.ai/solutions/ai-governance-platform.html): Permissions, named-person approvals, spending limits, and records that can be checked for integrity offline on actions routed through AAES Govern. Verification checks integrity; it does not establish independent attestation or complete capture. Enforcement requires control of the agent's credential path. Commercial overview with stated limits and evaluation path. - [Agentic AI security](https://aaes.ai/solutions/agentic-ai-security.html): Action-layer controls for enterprise agents: the threat model, what AAES Govern gates, and what it does not secure. Enforcement requires control of the agent's credential path. ## Guides - All guides: https://aaes.ai/guides.html - [AI agent access control](https://aaes.ai/guides/ai-agent-access-control.html): Identity, delegated scope, credential custody, request-bound human approvals, and a practical evaluation worksheet. - [MCP security checklist](https://aaes.ai/guides/mcp-security.html): Transport authorization versus business permission, credential handling, bypass and retry tests, and the documented AAES adapter limits. - [Enterprise AI governance platforms: a comparison](https://aaes.ai/guides/enterprise-ai-governance-platforms.html): Which platforms large organizations evaluate, what each layer governs, and the costs and control requirements smaller teams should evaluate. Includes FAQ answers on platform choice. - [Human-in-the-loop AI: examples and patterns](https://aaes.ai/guides/human-in-the-loop-ai.html): Training-time vs action-time loops, approval patterns, and when human-out-of-the-loop operation is acceptable. Includes FAQ answers on approval workflow platforms. - [AI governance platform: what to evaluate](https://aaes.ai/guides/ai-governance-platform.html): Model governance vs agent governance; agent access control, guardrails, monitoring, and audit trails. Includes FAQ answers on agent access control solutions. - [AI agent spending controls: patterns and tools](https://aaes.ai/guides/ai-agent-spending-controls.html): Cost observability, LLM gateways, and action-layer budget reservations; how configured ceilings limit authorization without capping final provider charges. - [AI agent control evaluation kit](https://aaes.ai/guides/agent-control-evaluation-kit.html): A reusable 14-test agent control checklist, procurement questions, and a runnable offline export exercise with recorded synthetic results. - [AI regulatory compliance: a practical guide](https://aaes.ai/guides/ai-regulatory-compliance.html): What compliance asks of teams running agents, what compliance software can and cannot do, and the limits of decision records as compliance evidence. Not legal advice. ## Contact Email hello@aaes.ai or use the contact form at https://aaes.ai/contact.html. Kubernetes is a standard supported install on GKE for AAES Operate and Govern, with backup and restore. The tested versions are GKE 1.36 and Kubernetes 1.37 in kind CI. It needs Helm 3.12 or later, Linux nodes with at least 8 CPUs each, encrypted persistent storage, a CNI that enforces NetworkPolicy, and ingress with TLS. The bundle needs a ReadWriteMany storage class. Customer identity and recovery checks are required before acceptance. Both can also be hosted by AAES.