01 / Connect
Bring your tools, data, and channels. No rip and replace.
Platform
The operating system for private AI agents.
Build, deploy, and operate purpose-built agents across your enterprise—inside your cloud, on your hardware, under your control.
02 / Govern
Scope every agent with explicit permissions, approvals, and policies.
03 / Deploy
Run specialized agents in infrastructure your organization controls.
How it works
From internal prototype to governed production system.
Aerie turns the prompts, scripts, and tools your team has already built into agents that can work reliably inside the systems you already run.
Connect
Integrate systems of record, internal tools, data stores, and communication channels.
/ connectShape
Turn existing prompts, scripts, and operating knowledge into purpose-built workflows.
/ shapeEvaluate
Test agent behavior against the business outcome—not just whether a response looks right.
/ evaluateDeploy
Place each agent in your chosen environment with its own context, skills, and permissions.
/ deployOperate
Monitor activity, route approvals, inspect records, and improve performance over time.
/ operateMulti-agent system
Purpose-built intelligence for every operational function.
Separate agents for separate jobs—scheduling, intake, coding, review, support, and the workflows unique to your business.
Gateway
One control plane for agents, users, channels, tools, events, schedules, and approvals.
Memory
Persistent operational context across conversations and time, scoped to the right agent and user.
Skills
Modular capabilities that let agents act through the systems your organization already trusts.
Infrastructure
Own your agents.
Aerie is for teams that cannot—and should not—send their most important workflows into a generic shared service.
Deploy where you say
Your cloud account or your own hardware. The agent runtime lives inside the boundary you nominate.
Route without lock-in
Use approved managed APIs or compatible open-weight models served inside your network.
Work where the work lives
Agents act through governed connections to your tools rather than forcing data into a new destination.
Control
Enterprise controls from the beginning.
The defining question is not only what an agent can do. It is what stops it, who approves it, and whether you can reconstruct what happened.
Human approval
Require a person before an agent writes to a system or performs a consequential action.
Sandboxed execution
Resource-limited workloads, defense-in-depth isolation, and enforced egress policies.
Scoped access
Give each agent only the systems, capabilities, and context required for its role.
Audit trail
Record actions, sessions, tool calls, approvals, and outcomes for inspection and export.
Product status
Built from a real production runtime.
Aerie is in development. The foundation operates in managed KiwiClaw today; customer-controlled deployment and enterprise-specific integrations are the next layer.
We would rather show the boundary clearly than let a landing page imply the entire system is already finished.
Running today
- Agent runtime and control plane
- Per-agent approval policies
- Sandboxing and egress controls
- Skills, memory, channels, and schedules
- Activity records and export
Being built with partners
- Customer-cloud and on-prem deployment
- Private model routing
- Systems-of-record integrations
- Continuous workflow evaluation
- Customer-owned commercial structure
KiwiClaw is not currently SOC 2 certified, does not currently sign HIPAA Business Associate Agreements, and does not currently provide a standard DPA. If those are gating requirements, we will address them directly during design-partner qualification.