Skip to main content
Epode runs at the company layer. The customer does not install Epode or create an account. The product foundation is to observe the AI channel and understand the customer journey. Conversion is the desired business outcome, but Epode does not yet assume one product mechanism for turning context into conversion. The agent experience graph is the core context surface:
Permissioned customer context remains available inside that loop:

Ask

The company integration requests useful context for one selected product interaction. Epode returns a bounded question/action tied to that interaction, identity state, use purpose, and expiry. HTTP agents answer through the company’s same-origin relay; MCP agents call tools registered by the company’s own MCP server. The company middleware waits only for this selected enrichment lookup, bounded by timeoutMs (250 ms by default). An Epode timeout or outage removes the optional question and returns the normal product response unchanged.

Learn

An accepted answer becomes typed customer context with provenance and consent. Epode stores only the bounded answer items—not the prompt, transcript, product response, or agent identity.

Personalize

The company’s server calls context.get with its own opaque identity references and one exact purpose. The result contains only active signals permitted for that purpose. Ephemeral context can be retrieved only for the same interaction. The company records which returned signal IDs affected a real decision. This prevents vague attribution.

Measure

The company reports a conversion, completion, engagement, dismissal, or abandonment linked to that decision. Identical retries are idempotent; conflicting retries fail closed.

Existing outcome feedback

Structured product feedback remains supported and supplies useful outcome/friction evidence. It is no longer the primary onboarding path. Customer enrichment and outcome feedback are supporting context and evidence capabilities; neither defines a proven conversion mechanism.