Docs / Beyond investing
Beyond investing
Build an agent people can check before they trust it. Epoche brings confidential code review and verifiable deployments to private apps, general agent services, and tools that act on someone else's behalf.
The same foundation, different applications
An investment strategy needs evidence about where it can send money. A private document assistant needs evidence about where it can send data. Both start with the same question: does the code follow the rules its users rely on?
The approach behind Epoche's investment products applies to both. Submit code for confidential review, request the certifications that fit the application, then connect those review results to the code that runs. Your users can check the evidence without needing access to your source.
An agent can carry several certifications. Each covers its own rules and records what the review examined. The trust chain connects that evidence to the running agent; What it proves explains the scope of each claim.
Start with the reviews available today
General agent services
Build a research assistant, document converter, or workflow service that another person or agent can call. Execution Integrity reviews declared capabilities, dependencies, network destinations, and secret handling. It gives callers evidence about how the service executes, while they still judge the quality of its work.
Apps with verifiable privacy claims
Build a private document assistant or a service that processes sensitive inputs. Privacy Integrity reviews how the code handles those inputs, including logs, storage, and outbound destinations. A confidential runtime and a live check help connect those reviewed rules to the service a user actually reaches.
Name the privacy claim precisely. For example, whether input reaches application logs is a question a review can examine. Any external services that receive data remain part of the privacy assessment.
These are the existing Execution Integrity and Privacy Integrity domains. See their review rules and the CLI guide to start building.
More questions an agent could answer
Could a service check a professional credential before granting access? Could an agent prove it needs your approval before sending an email? Could a team inspect how an app handles health records without seeing its proprietary code?
Below are proposed certification domains for that broader platform. All are marked Coming soon: they are directions for future reviews, not certifications you can request today. Names and scopes may change; no release dates are set. Each would need its own published rules, tests, and evidence requirements.
Data and privacy
- Consent and purposeComing soon
- Check that an app requests permission for a named use of data and checks that permission before processing it.
- Retention and deletionComing soon
- Review when an app stores data, how long it keeps it, and whether deletion reaches the stores it controls.
- Data residencyComing soon
- Check declared storage and processing destinations against a policy that names permitted regions and providers.
- Tenant isolationComing soon
- Review whether one customer can read or change another customer's records, files, or agent memory.
Identity and credentials
- Credential verificationComing soon
- Check issuer signatures, expiry, revocation, and the connection between a credential and the person or agent presenting it.
- Access and eligibilityComing soon
- Review whether every protected action checks the required role, membership, age, qualification, or other credential.
- Selective disclosureComing soon
- Check that an app requests and reveals only the credential fields needed for a particular decision.
- Delegated authorityComing soon
- Review how an agent receives permission to act for someone, how that permission is limited, and how it can be revoked.
Agent actions
- Human approvalComing soon
- Check that actions such as sending a message, publishing content, or deleting a record wait for the required approval.
- Tool permissionsComing soon
- Review which tools an agent can call and whether each call stays within the permissions granted for the task.
- Usage and cost limitsComing soon
- Check that API calls, compute use, and other metered actions respect declared budgets, including retries and restarts.
- Untrusted input handlingComing soon
- Review how documents, web pages, and tool results reach instructions and privileged actions. Define specific attack cases to test.
Sensitive workflows and records
- Health-data controlsComing soon
- Review access, disclosure, and audit controls for apps handling health information, with a defined scope for each review.
- Confidential business workflowsComing soon
- Check rules for handling contracts, employee records, legal documents, and other restricted business data.
- Audit recordsComing soon
- Review whether required actions produce attributable records, whether events can be omitted, and how records avoid exposing private inputs.
- Output provenanceComing soon
- Check how an app records the sources, model or tool versions, and processing steps behind a result. Provenance alone does not establish accuracy.
What this could mean for healthcare
A patient intake assistant or medical document processor could carry evidence about specific health-data controls: who can access a record, which services receive it, and what the app logs. That is the goal of the proposed health-data review.
HIPAA compliance also depends on the organization's practices, agreements, and safeguards outside the code. A code-review credential would support that work, not certify the whole app or organization as HIPAA compliant. HHS explains the broader requirements in its Security Rule summary.
What would you build with it?
Start with a promise your users need to check. Choose the review rules that could support it, and make the evidence available to whoever relies on your service. Epoche Build will be home to this broader developer platform.