Skip to content
How Optophi works

One boundary for agent action.

Give agents defined tools, apply policy before execution, inspect the outcome, and make multi-step changes through explicit state.

The public model

Tools. Policy. Audit. State.

These four parts give a team a simple way to reason about what an agent can discover, what it can execute, and how to inspect a decision.

Tools

Describe operations with input schemas. Show each agent the subset selected by its tool profile.

Policy

Validate and apply configured rules before a side-effecting adapter runs. Return a clear refusal when a call is denied.

Audit

Write structured records for operations and decisions when an audit writer is configured.

State

Define legal transitions for longer tasks and move work through explicit states.

The working path

Try the boundary locally.

The available local MCP server exposes a selected operation profile. You can run a simulated operation, configure policy, and connect an audit writer through the embedding runtime.

  • Scope the catalogChoose the operations visible to the agent.
  • Check before actingValidation and configured policy run at the operation boundary.
  • Inspect the resultSee the adapter used and whether a result was simulated.
Run the local MCP quickstart
The hosted path

Connect once.
Keep the boundary.

Provisioned hosted access uses an API-issued key bound to a tenant and tool profile, shared invocation through MCP and REST, and audit IDs for tool calls.

Current release: Optophi API access. Create a scoped validation key through Optophi onboarding and run a guided first task over MCP or REST. Existing provisioned keys remain supported. Local examples require no account. Connect with your provisioned key.
Read about trust boundaries
Feedback