xpander.ai, a startup founded by three former Amazon Web Services (AWS) principal engineers, has made its enterprise AI agent platform generally available and announced a $7.5 million seed round. The company positions its product as a vendor-neutral control plane that sits above models and agent frameworks to handle execution, permissions, observability, memory and lifecycle management across heterogeneous environments.
Why the layer matters
Industry estimates highlight rapid agent proliferation: Gartner expects the average Fortune 500 company to be running more than 150,000 AI agents by 2028, up from fewer than 15 in 2025. At the same time, only 13% of organizations say they have appropriate AI agent governance today. That gap creates demand for infrastructure that standardizes operations without requiring each new agent to rebuild the same supporting services.
What xpander.ai offers
At the heart of xpander's platform is the "Universal Harness," described as a model-, framework- and cloud-agnostic runtime for executing agents as portable enterprise workloads. Customers can use xpander's hosted environment or self-deploy the platform on Kubernetes or on-premises under the enterprise offering. The company names AWS, Google Cloud, Microsoft Azure, private VPCs and fully air-gapped on-premises environments as supported deployment targets.
xpander says it accepts agents built with frameworks such as LangChain, Strands and Agno, and supports proprietary, open-weight and customer fine-tuned models. Integration surfaces include a language-agnostic REST API for control-plane operations, a Python SDK for building agents and workflows, and Model Context Protocol (MCP) support to expose agents and tools to MCP clients.
The control plane provides centralized identity, audit trails, tool-call logging, human approvals, task-level spending attribution and credential injection from a vault (so credentials are not exposed directly to models). The company states it is SOC 2 Type II certified and GDPR compliant; the enterprise tier adds SSO/OIDC, a private model gateway and sub-organization reporting.
Vendor neutrality — and the new dependency problem
xpander emphasizes vendor neutrality: enterprises can swap models, frameworks and underlying infrastructure. However, xpander's Universal Harness and control plane become the coordinating layer, which could create another form of lock-in if operational state and configurations are hard to migrate to a different control plane. Public documentation does not yet detail portability options if an enterprise ends its license.
A crowded, evolving market
The market has evolved beyond a binary choice between hyperscaler platforms and neutral alternatives. Competitors and adjacent solutions include:
- LangChain (LangSmith): deployment and governance for production agents with self-hosted and hybrid options.
- CrewAI: centralized governance, SSO/RBAC, support for cloud, VPC or customer-managed infrastructure, and multi-model support.
- Temporal: durable execution for long-running AI workflows (retries, recovery, human approvals) rather than a full agent management suite.
- Major vendors: OpenAI (Frontier) and Google (Gemini Enterprise Agent Platform) are building upward-facing services that add permissions, execution and governance around agents.
Because many vendors already support multi-model and customer-controlled deployments, xpander's claim must be judged on how well it combines framework and model portability with enterprise identity, governance, runtime features and collaboration.
Collaboration and "Multiplayer AI"
xpander is also pushing a collaboration layer it calls "Multiplayer AI." The idea is to turn agents into organizational assets rather than personal assistants: shared, persistent conversation threads and permission-scoped workflows that outlive individual chat sessions. Agents can be published for organizational use and integrated with Slack, Teams, ChatGPT, Claude and xpander's own UI. The platform supports end-to-end authentication via OIDC so downstream actions can be attributed to the human who invoked an agent.
Omni: a prebuilt agent and benchmark results
xpander is also making Omni, its prebuilt, general-purpose agent, generally available. Omni is described as an "AI forward-deployed engineer" that helps turn a requested business outcome into an "Agentic Application": a backend agent plus frontend experience (chat, UIs, reports, dashboards). Omni is intended to assemble requirements, build the backend agent, attach connectors and generate a live application surface.
xpander reports Omni scored 90.9% on the GAIA benchmark, combining multiple model families during testing. The company presents this as evidence that agent performance depends heavily on the surrounding harness and runtime services, though the score is company-reported and not an independent validation.
Pricing and availability
xpander's platform and Omni are generally available now with two main commercial models:
-
Team (hosted): usage-based credits with no seat charges. One credit equals $0.01. Each event or message that wakes an agent costs one credit for the entire turn; each tool or API call costs an additional credit. Model token usage is billed separately according to configured model rates. New accounts receive 1,000 free credits. The Team tier allows unlimited agents, workflows and seats.
-
Enterprise (self-hosted): annual license starting at a 50-agent commitment, supporting Kubernetes or on-premises deployments, SSO/OIDC, private model gateway, sub-organization pooled credits and Tier 1 support. The company does not publish a public dollar price for the annual enterprise license; enterprise customers must contact xpander for custom pricing.
Under the hosted model, xpander markets "you pay only for the work the agent is doing" with no subscription or seat fees; self-hosting customers must consider the undisclosed license plus infrastructure and model-provider costs.
Where xpander fits and the questions ahead
xpander's pitch is straightforward: let employees keep using the models, frameworks and interfaces they prefer, but move control (permissions, monitoring, execution) into a common layer the organization governs. Its success depends on proving that an independent control plane adds enough value to justify another platform in an already crowded enterprise AI stack, and on demonstrating migration options, scale, and production reliability in real deployments.
The coming months will show whether enterprises favor a neutral vendor like xpander for that control plane, extend agent management from model and cloud providers themselves, or consolidate around agent-framework vendors and durable-execution platforms.
