Agentic API Orchestration

ReṼoman is the middleware between an agent and your API-Graphs. Hand it an API-Graph; it executes that graph deterministically on the JVM and hands back the Rundown — the Context Information the agent reasons over.

Under the hood, the mechanism is an API-orchestration engine: you hand it a workflow of interdependent API calls as executable API metadata — an API-Graph — and it executes that graph. That precise verb is what keeps the middleware headline honest: the engine runs a given graph; it does not decide which graph to run. The agent does the orchestrating — choosing and coupling graphs; ReṼoman is the execution step it delegates to.

Who the agent is

"Agent" here is any caller that reasons over the Rundown to decide what runs next — an AI-agent Node, or your own JVM/test code acting as one. When your JVM code feeds ReṼoman an API-Graph and asserts on the returned Rundown, you are the agent; the same substrate serves an AI-agent Node. This is why the middleware role is real today, not aspirational: the caller port ships now, and the AI-agent conveniences (see Roadmap) are glue on top of it.

The loop

The building block ReṼoman provides for agentic systems is a deterministic execution step that takes executable API metadata in and produces Context Information out:

// ORCHESTRATION // EXECUTION Agentic Node chooses & couples graphs reasons over Context Information ReṼoman executes the graph deterministically · on the JVM API-Graph Rundown = Context Information

  1. executable API metadata (API-Graph) — the workflow of interdependent calls, with the order and the data dependencies between them.

  2. ReṼoman executes the graph — deterministically, on the JVM: resolving data dependencies between calls, handling auth, polling for async outcomes, and marshalling type-safe payloads.

  3. Rundown / toJson(VERBOSITY) — a structured record of everything that happened. This is the patent’s Context Information.

  4. An agentic-workflow Node consumes that Context Information to decide what comes next.

This ships today, not aspiration

The output an agentic Node consumes is the real shipped Rundown. Calling Rundown.toJson(Verbosity.VERBOSE) emits stepReports / mutableEnv / stats — the exact shape a Node reads:

  • stepReports — per-step request/response, marshalled payloads, and outcome.

  • mutableEnv — the environment as mutated across the run, carrying values resolved between steps.

  • stats — execution statistics and timing.

This is a proof point that exists in the engine today, not a promise. See Rundown for the full shape.

A light analogy

Picture a restaurant kitchen. The API-Graph is the pre-planned menu set — the dishes and the order they go out in. ReṼoman is the kitchen manager that directs the line cooks to actually produce those dishes, then hands the head chef a tidy summary of what was served and how it went. The manager executes the menu it was given; it does not invent the menu.

Execution vs. orchestration: staying precise

The patent draws a clean line:

  • Orchestration (the agentic layer) chooses and couples API-Graphs to workflow Nodes — deciding which graph runs and how its output feeds the next decision.

  • Execution (ReṼoman) runs a given API-Graph and returns Context Information.

ReṼoman is strictly the execution engine. The product headline — middleware between an agent and your API-Graphs — is held precise by the verb "executes the graph": ReṼoman is the deterministic execution layer beneath an agentic orchestration system, not the agentic layer itself.

Roadmap

The following are direction — where this is heading — and are not shipped today. Do not read them as present-tense capabilities:

  • MCP tool-definition autogen — generating MCP tool definitions from OAS plus API-Graph metadata, so an agent can discover and call a graph as a tool.

  • Dynamic coupling of agentic workflows to API-Graphs — letting an agentic layer bind workflow Nodes to API-Graphs at runtime.

Today, ReṼoman delivers the deterministic execution half of that picture: feed it an API-Graph, get back Context Information.