Today, we’re announcing xpander’s $7.5 million Seed round led by Pico Venture Partners, with participation from Emerge Ventures, Samsung Next, and SeedIL, alongside the general availability of the xpander enterprise AI agent platform.
Enterprise AI adoption is now nearly universal, but mature deployment is still rare. Most organizations fail not because employees lack interest in AI. It’s because agentic work is happening outside the infrastructure the business can govern, scale, and trust. It’s happening inside employees’ laptops.
The answer isn’t stopping your teams from using agents. It’s giving them a platform that makes agents safe, governed, and auditable.
xpander moves agentic work off unmanaged laptops and into governed, vendor-neutral, multiplayer infrastructure that every company can own, manage, and audit.
Desktop agents are running wild
Employees are already using tools like Claude Desktop, Codex, and other agentic apps to get work done.
Although the enthusiasm is good, ungoverned agent usage creates serious enterprise risk. When agents run locally on employees’ laptops, companies lose visibility into:
what each agent accessed
which files, tools, and APIs it used
who authorized a given action
what the agent changed, sent, summarized, or produced
whether the workflow can be reused by anyone else
The result is agentic work that is powerful, but unmonitored. Productive, but unaudited. Useful to one employee, but siloed from the rest of the company.
We’ve heard from enterprise customers that they have to block powerful desktop agent apps like Claude Cowork, even when employees love them. These tools can use an employee’s computer to access internal systems, and that makes agentic behavior too risky for security and compliance. xpander solves this by giving employees the same capabilities, but in a governed, monitored, auditable environment the business controls.
Until today, enterprises had two bad options
Most companies that wanted to operationalize AI agents had to choose between two imperfect paths.
Option one: go all in with one vendor
This may look like a single solution, but it’s actually not. Single-vendor services give you model endpoints, basic tool calling frameworks, and authentication APIs. This isn’t a complete, usable system.
To turn those primitives into something employees can actually use, you still need a platform engineering team to build the interfaces, governance, and workflows that make agents work in your business. You’re getting the building blocks, but still need to scaffold them into a usable platform.
And those building blocks still come with lock-in. Your workflows, prompts, skills, and agent behavior become dependent on one vendor’s roadmap, pricing, limits, and interface. If a better model emerges, a policy changes, or your deployment requirements shift, your work is trapped.
Option two: gluing together multiple pieces of different solutions
This gives teams more flexibility, but creates a different problem. You need one tool for model access, plus separate systems for orchestration, tool calling, sandboxing, governance, observability, and collaboration.
That stack can work for a small technical team. It doesn’t become a durable operating layer for the whole enterprise.
Either way, you shouldn’t have to choose between vendor lock-in and infrastructure sprawl.
Today, companies do not need to choose anymore
xpander gives enterprises a third path: a governed platform for agentic work that is multi-vendor by default, deployed by the company, built for teams, and actually covers the end-to-end capabilities you need to get value from AI.
With xpander, you can:
Move agents off unmanaged laptops and into controlled infrastructure
Build, deploy, and govern AI agents across any model, cloud, or framework
Use Claude, GPT, Gemini, open-source models, or the company’s own fine-tuned models through one platform
Run agents in the cloud, private VPC, or even air-gapped environments
Create audit trails for what agents did, what tools they used, and what data they touched
Share prompts, skills, workflows, and agents across teams instead of leaving them trapped with one person
Use Omni as an out-of-the-box enterprise agent and agentic Forward Deployed Engineer to turn complex requests into production-grade agents and workflows
Let employees work through Slack, Microsoft Teams, desktop AI apps, embedded product experiences, or a shared multiplayer AI user interface
This is how enterprises become AI-native without giving up control over security, compliance, or governance.
xpander gives every company one governed home for AI agents
Under the hood, xpander is built around its universal harness, a unified runtime that lets you run reliable, governed agents with any AI model and in any environment.
The harness connects to Claude, GPT, Gemini, open-source models, or your own fine-tuned models through a single model gateway. It runs agents in infrastructure you control, whether it’s cloud, private VPC, or air-gapped. You’ll get full audit trails of agents’ actions, tool calls, and data access.
For long-running workflows, the harness handles the hard infrastructure: tool calling, sandboxed code execution, persistent memory, and automatic recovery when tasks drift or fail. This is what lets agents run for days instead of minutes without falling apart.
On top of the harness, xpander provides the platform layer: Omni as your agentic Forward Deployed Engineer, shared prompts and skills across teams, and integrations with Slack, Teams, desktop AI apps, and a multiplayer UI.
Together, the harness and platform give enterprises one governed home for agentic work, where agents are safe, shareable, and scalable across the business.
What it means to become AI-native
Every company wants to become AI-native. But being AI-native doesn’t mean letting every employee bring their own agent to work, run it locally, and hope the business can make sense of the results later.
Being AI-native means giving teams shared infrastructure where agents can be governed, audited, improved, and reused across the business.
For us, that comes down to five pillars:
Multi-vendor by default: Companies should be able to use Claude, GPT, Gemini, open-source models, or their own models interchangeably. AI should not create a new version of vendor lock-in.
Owned by the enterprise: Agents should run in infrastructure the company controls, with clear rules for what they can access and what they can do.
Orchestrated beyond the model: Models are becoming more interchangeable. The real advantage comes from how agents plan, call tools, stay focused, check their work, and complete multi-step workflows.
Built for multiplayer work: Real enterprise work is not one person chatting with one bot. It is multi-day, multi-person, and often multi-agent. AI infrastructure should make that collaboration possible.
Accessible to every company: The infrastructure to run agents properly should not be reserved for the Fortune 50. Every company should be able to become AI-native without building the entire stack from scratch.
This is what we mean by AI-native transformation: not adding another chatbot to your stack, but rebuilding how work gets done around governed, multi-vendor, multiplayer agents.
Additional read:
The proof: 90.9% on the GAIA benchmark
Governance can’t come at the expense of capability. If agents are going to move from laptops into enterprise infrastructure, they still need to perform.
At xpander, we’re not theorizing about agents. We build them, run them, and use them every day. We know where they break: tool use, context, long-running tasks, recovery, evaluation, and keeping the agent focused until the work is done.
On the GAIA benchmark, which tests how well agents handle real, multi-step tasks, xpander scored 90.9%, surpassing proprietary, fine-tuned solutions built specifically for the test.
The result confirms what we’ve learned from building real agents: the model matters, but it is not enough. The real gains come from the infrastructure around it: the runtime, tool use, sandboxing, focus, and feedback loops that help agents complete work reliably.
What’s next
This funding helps us move faster on the infrastructure every company needs to make agentic work governable, shareable, and scalable: stronger governance, broader model support, and easier ways for domain experts to build agents that teams can trust and reuse.
The company that leads the next phase of AI adoption won’t be the one with the most desktop agents. It will be the one that can govern, share, and scale agentic work across the business.
If your team is already hitting the limits of unmanaged desktop agents, you don’t need another disconnected AI tool. You need a governed, multi-vendor, multiplayer way to run agents across your business.
See xpander in your own environment
Talk to our team about enterprise deployment, onboarding, and support options.
Book a demo
