Available now
Agent plugin
Inside the harness you already use, around each tool call. A harness is the software that runs the model and calls tools. This is the quickest place to start.
The product
The Compliance Operator - CECO
It allows an agent to make a tool call, or blocks it, before the tool runs. The policy you state is checked in mathematics and written into code. The gate checks the call and its inputs against that code. An audit record keeps what was allowed, what was declined, and the result. The policy version, the proposed call, and the result form a secure compliance passport.
You choose where that gate sits. Each place is a form factor of the same product, and each one runs the same policy. An agent, ordinary software, and a person all have to pass it. A later form factor reuses the policy you already wrote.
Available now
API or protocol control layer
In front of one system of record, on its API or on a protocol it already speaks. CECO holds the credentials, so that system accepts a call only from the gate. We build this with you.
April 2027
Container sidecar
Beside each service, with the network sending traffic through the gate. For teams running agents on a shared platform.
Research
Proved FPGA programming
The same policy is compiled into the program an FPGA runs, and that program is proved against the policy, down to the circuit. The chip sits outside the caller’s host, or on the only route to the protected system. There is no date yet.
SMART: SCOTLAND
Funding for the framework
Scottish Enterprise awarded Riley Betts SMART: SCOTLAND funding for the Compliance Operator Framework. The project runs from 18 June 2026 to 30 June 2027. It furthers the framework around the Compliance Operator. It is not another name for the product.
Working with a firm
Write the policy, then prove it
Riley Betts works with your firm to write the agentic execution policy and the authority policy. Spec++ is how that policy is written. Lean is how it is proved and implemented as a versioned control.
Roadmap
Anchor Proof Framework
Anchor uses the same formal-methods pattern as the Compliance Operator. The gate still decides. Anchor adds a proved policy for how agents negotiate and interact with each other. That policy sits above the agent protocols already in use. It does not replace them. Anchor also adds a receipt a third party can check: which policy version, which proposed call, and which result, without the private inputs. The receipt work extends that pattern onto a public or private chain and into lean-tee, a trusted execution environment implemented as a zkVM (zero knowledge virtual machine). It is on the roadmap. It is not a second product you can adopt today.
Use cases
Patterns researched and explored
The same shape has been explored in a few settings. An agent proposes a change to a record, and the gate allows or declines it before the change. Two firms confirm a trade under a shared policy, and the private book stays off the kept record. A supplier status change is checked against a pinned policy before it runs. Each one is formal methods in support of a Compliance Operator, and a step toward high-assurance agentic adoption.