Chorigen Governed AI-App Platform

Desktop Hub acts as a local trust boundary and checkpoint on employee machines. It ensures that AI apps run within a governed environment where every tool call, credential use, and network request is intercepted and enforced at the execution boundary. Key security features include:

• Deterministic Enforcement: Access is controlled by compiled policy rather than intent-based filters.
• Credential Protection: Raw credentials are held by the checkpoint and are never exposed to an agent's context window.
• Local Execution: Apps run on the device via a runtime adapter, allowing for offline operation (though a connection is required to browse the catalog or install new apps).

Cloud Studio utilizes two distinct design planes within a single space:

1. The Workflow Plane: Used to draw the overall process, defining who performs which action and where human sign-offs are required. A finished agent is inserted into this plane as an 'agentic task' alongside human and service participants.
2. The Agent Plane: Used to design the internal logic of the agent, specifying exactly which tools it can access and how it reasons through tasks.

Cloud Studio allows owners, analysts, and developers to co-design multi-party workflows using BPMN-inspired diagrams. The design process consists of two planes:

1. Agent Design: Define model steps, tool calls, human checkpoints, and a bounded reason-act loop. The model is restricted to only the tools allowed by the design.
2. Workflow Integration: Place the designed agent into a workflow as an 'agentic task' alongside human and service participants.

These designs are processed by a model-to-app compiler that turns diagrams into role-specific apps, binding guardrails directly into the app at compile time to ensure the app cannot perform actions not declared in the design.

Desktop Hub (v0.9.1) is available for Windows, macOS, and Linux. Because it is a pre-release build and not yet code-signed, users will encounter OS security warnings during installation:

• Windows: SmartScreen will warn; select 'More info' then 'Run anyway'.
• macOS: Gatekeeper will warn; open the application from Finder and select 'Open anyway'.

Update Behavior:

• Windows and Linux: Update automatically in place.
• macOS: Requires manual download of new builds.

Chorigen uses phone numbers to send SMS messages for two-factor authentication (2FA), screening reminders (daily/monthly/yearly), and administrator notifications regarding new member registrations.

• Opt-in: Users can opt in by providing their phone number and agreeing to the privacy policy via the user dropdown menu in the web application.
• Opt-out: Users can opt out by deleting their phone number from the application.

Cloud Studio provides a client-side sandbox for testing agent designs without requiring API keys, network connectivity, or model spend. Key features include:

• Manual Model Simulation: The designer 'plays the model' to test logic flow.
• Fixture-based Tools: Tools return predefined fixture data instead of making live calls.
• Admission Checks: The sandbox refuses to run designs with open/incomplete scopes before execution begins.
• Determinism: Recorded episodes can be replayed exactly to ensure consistent behavior.

The Cloud Studio interactive demo provides a full-page designer environment with pre-loaded sample designs. Users can interact with the canvas to move tasks, edit forms, and trace expressions.

Key constraints of the demo:

• No Account Required: There is no sign-up or credentialing process.
• Local Storage: All changes are stored locally in the browser's storage on the user's device. Clearing browser storage deletes all modifications.
• Design Only: The demo is limited to the design phase. It cannot compile designs into apps, publish them, or run them, as these actions require a real workspace.
• Isolation: The demo does not connect to real Chorigen workspaces or servers; no data is sent to the provider.

In Cloud Studio, agents are constructed as designs rather than text prompts. They are built using a fixed vocabulary of components:

• Model steps: Basic LLM processing steps.
• Tool calls: Specific functions the agent can execute.
• Ask-user checkpoints: Points where a human must make a decision.
• Reason-act loops: Bounded arenas where the model can choose from a specific set of allowed tools.

Crucially, the behavior of a tool call changes based on its placement: a tool call outside a loop is a structural requirement (it always happens), while a tool call inside a loop is a candidate the model may choose to select.

The Desktop Hub serves as the local trust boundary for employees. It provides a centralized location to discover and run IT-approved, checksum-verified apps.

Key security features include:

• Execution Boundary Enforcement: Every tool call, credential use, and network request passes through a hub-controlled checkpoint beneath the model layer.
• Local Execution: Apps run locally (live or offline), keeping personal traffic on the device.
• Policy Enforcement: The hub runs apps according to the policies bound to them during the compilation process.

The Chorigen workflow follows a three-step process:

1. Design: Use Cloud Studio to co-design multi-party workflows using BPMN-inspired diagrams. This includes defining AI agents with model steps, tool calls, human checkpoints, and bounded reason-act loops.
2. Compile: The model-to-app compiler converts the design into role-specific apps with security policies baked in at compile time.
3. Run: Compiled apps are deployed to the Desktop Hub, where employees can discover and run them (online or offline) under IT governance.