Accepting design partners

The agent firewall for the enterprise.

Observability and kernel-grade isolation for every AI agent — sanctioned or not — running on your devices. Deploys alongside your EDR. Built for the agentic enterprise.

No spam. We'll reach out to schedule a demo.

Thanks. We'll be in touch shortly to schedule your demo.
The problem

Your EDR can't see this.

Every enterprise is deploying AI agents. None of them have a way to control what those agents actually do once they're running on a laptop.

88%
of organizations reported an AI agent security incident this year.
+$670K
average shadow-AI breach cost above the cyber breach baseline.
12–18
month window before this category has an incumbent. We're shipping now.

First-generation AI security — prompt guardrails, output classifiers — was built for chatbots. It's blind to runtime agent behavior on the endpoint. Agents now execute code, touch filesystems, write to the registry, open network connections, and act across production environments. EDR doesn't see them as threats. AI security tools don't see the OS-level actions.

And the perimeter just collapsed. Anthropic's Claude Dispatch lets a phone control a desktop computer remotely — clicking buttons, monitoring programs, auto-approving actions. Every other major lab is racing to ship the same capability. Without an agent firewall, the enterprise has no way to enforce what the agent can do, where it can reach, or whether it's even authorized to be there.

What Ospiri does

Observe. Sandbox. Govern.

A single endpoint agent and Windows kernel driver. Deploys alongside your EDR stack and intercepts agent activity at the OS layer — before the action reaches the filesystem, the registry, the network, or another process.

Observe

See every AI agent on every endpoint — sanctioned, unsanctioned, or installed by an employee five minutes ago. The signature pipeline catalogs Claude, ChatGPT desktop, Cursor, Copilot, Gemini, and emerging tools surfacing on the dark web.

Sandbox

Copy-on-write isolation across four kernel scopes. When an agent tries to modify a file, Ospiri clones it into a sandbox rather than blocking the write. The agent gets the functionality it needs. The original files stay untouched.

Govern

Policy-driven isolation that enables governed agent productivity instead of breaking it. Decide whether to commit, discard, or escalate sandboxed changes. Unknown agents get automatic restrictive policies pending admin review.

Kernel-grade isolation

Four layers of control. One driver.

The kernel driver enforces policy across four scopes — with a fifth in active development. Each layer is enforced before the action reaches the OS.

01

File system isolation

Controls which paths an agent can read, write, modify, or delete. Prevents unauthorized access to network shares, sensitive directories, and customer data.

02

Registry isolation

Controls Windows registry writes. Stops agents from establishing persistence, modifying system settings, or tampering with other software.

03

Network isolation

Per-process firewall built on the Windows Filtering Platform. Block a coding agent from reaching Salesforce, allow it to reach GitHub. Granular, policy-driven.

04

Object isolation

Controls inter-process communication. Stops agents from injecting into or coordinating with other processes on the box.

05

WSL & VM monitoring Roadmap

Extends the same isolation guarantees to Windows Subsystem for Linux containers and VM-resident agents.

The architectural bet

Copy-on-write, not block-on-deny.

The difference between an agent firewall that breaks agent productivity and one that enables governed agent productivity. Built where the OS is decidable: at kernel scope.

The signature pipeline

A continuous research system catalogs AI agent binaries — installers, executables, registry keys, network behaviors. Coverage of known agents and emerging tools surfacing on the dark web. The same architectural pattern as endpoint AV signature distribution.

The technical moat

Kernel driver work that integrates with undocumented Windows internals. Copy-on-write semantics across four kernel scopes is the part that's hardest to copy — we estimate 12–24 months for a competing team to build equivalent isolation.

Who it's for

One platform. Three ways in.

Ospiri meets you where your agent risk lives — whether you're a CISO standardizing across thousands of endpoints, a small team that just wants visibility, or a partner embedding governance into client deployments.

The team

Operators who've shipped enterprise security at scale.

Yasir Ali

Co-founder & CEO

Former CEO, Polymer Data Security (founded 2020, acquired by Acacia Group 2026). Former developer and mortgage quant at Bear Stearns; later a structured products trader and portfolio manager at JWM Partners and the Barclays prop desk. Built a consultancy practice around data governance and technology helping organizations such as Deutsche Bank, Voya Financial, and CIBC modernize their stacks. Built and scaled Polymer — one of the first DSPM/DLP platforms for SaaS — with customers including Robinhood, RSA, Scale AI, Edward Jones, and CVS, and raised $8MM from prominent VCs. Brings a deep network across hedge funds and financial services.

Aaron Bray

Co-founder & CTO

Former CEO, Phylum (acquired by Veracode 2025). 14+ years in software engineering and information security, including 11 years inside the U.S. Intelligence Community and the U.S. Air Force before leading Sony's Global Threat Emulation red-team development. Co-founded and served as CEO of Phylum, a software supply chain security platform analyzing open-source packages across five classes of risk, which raised $19.5MM from top-tier VCs including ClearSky and Atlassian. Deep research background in malware diversity, software anomaly detection, program synthesis, and applying NLP to binary analysis — a rare attacker-mindset perspective on securing modern AI and agentic runtimes.

"The osprey doesn't watch the surface — it sees through it. Ospiri does the same for autonomous AI on your endpoints."
Accepting design partners

See every agent. Govern every action.

Join the waitlist for a demo. We'll reach out to schedule a working session and scope a deployment for your environment.

For enterprise and partner inquiries, mention your team in the demo conversation.

Thanks. We'll be in touch shortly to schedule your demo.