Countersign by FohBoh.ai is the MGE-powered trust layer that approves or denies agent actions before they execute. An AI agent calls Countersign before it acts. Countersign evaluates the proposed action against facts MGE has certified and authority your customer has registered, and answers in the same call: APPROVE or DENY, with the reason. Countersign never performs the action itself.
Countersign sits in the tool execution path, at the point where an agent's proposed action would become a real one. One call, one answer.
The agent proposes an action, such as paying an invoice, with its exact parameters.
Countersign reads what MGE has certified and runs eight deterministic Execution Controls: the evidence, the agent's identity and delegated authority, its limits and your rules.
APPROVE or DENY, in the same call. A DENY carries a reason code, a hint on what would change the answer, and a plain-language narrative. An APPROVE carries a single-use Execution Certificate.
The enforcing adapter in front of your payment system verifies and redeems the certificate before anything runs, and records what actually happened.
An agent that skips Countersign is refused at the destination. Enforcement lives outside the agent.
The Execution Certificate covers the specific action and parameters evaluated, and can be redeemed once.
No error, timeout or unavailable dependency ever results in APPROVE. A timeout is never a yes.
Limits, approvers and policies are registered in advance. Nothing an agent writes in free text can change a decision.
The Trust Score of the certified inputs must clear its threshold. It is necessary and never sufficient on its own, and Countersign produces no score of its own.
Observability captures the proposal and the tool arguments. Countersign answers four questions before any money moves. If required evidence is missing or a rule fails, the answer is DENY.
Is the invoice supported by evidence MGE has certified?
Is the vendor authorized?
Is the payment within the agent's delegated authority?
Has this invoice already been paid?
Above the agent's limit, the payment needs a sign-off from an approver your customer has registered, made with that person's own credential, never an agent's key.
Observability can run throughout an agent's work, including before execution. Countersign's defining role is the enforced checkpoint. The two work together.
Execution Authorization & Certification belongs directly in the tool execution path, connected to tracing, logging, alerting and audit dashboards. It is not a governance report at the end of the line; it is the decision an action has to pass before it runs.
That makes observability platforms a natural partner. Countersign contributes the enforceable decision. Their traces and dashboards make that decision, and the outcome that follows, visible.
Partner with us →Our customer is the agent producer: a software company whose agent acts inside its customers' systems, such as an accounts-payable, procurement or claims platform. You hold the keys and the contract. Your customers never hold a FohBoh key; you register their agents, limits, approvers and policies through an administration API, and each customer is isolated from every other.
A proposed action in, APPROVE or DENY out, synchronously, in well under a second.
Reason codes, a hint on what would change the answer, and a narrative the agent can relay to a person.
An MCP server and SDKs, so an agent on any framework can call Countersign like any other tool.
Every call is idempotent. A repeated request returns the same decision.
Every use case has the same shape: an agent proposes an action, MGE has certified the facts behind it, and Countersign decides before anything moves.
Stops: overbilled amounts, duplicates, a fraudulent bank change, payments above the agent's limit.
First use caseStops: paying more than is owed, paying twice, paying a new account without verification.
Stops: orders above authority, unapproved suppliers, prices outside the contract.
Stops: settlements above authority, payment to someone other than the claimant, denials without a person.
Stops: refunds above what was overpaid, write-offs above authority. Patient data stays out.
Stops: refunds outside policy, discounts above a set rate, repeat refunds to one customer.
An APPROVE means the action satisfied the configured Execution Controls on the available evidence. It is not a guarantee that every approved action will produce the business outcome you want.
Countersign is in development and not yet available. If you build AI agents that act for your customers, tell us which actions you need governed first.