@intelligo-dev/mastra

Install

pnpm add @intelligo-dev/mastra

@mastra/core 1.x is an optional peer dependency and nothing here imports it — the agent is typed structurally, so the bridge stays removable and the agent stays native. What this package adds is the recording.

Use

import { runWithExecution } from "@intelligo-dev/mastra";
import { executions } from "@/lib/intelligo";

const result = await runWithExecution(
  { executions, workspaceId, userId, capability: "support.reply" },
  () => supportAgent.generate(messages) // native Mastra, untouched
);

runWithExecution opens an execution, runs the agent and settles what it cost, reading usage and the model from whatever shape the provider reported. executions is the instance the composition root builds with createExecutions from @intelligo-dev/executions; naming the model sizes the credit hold admission takes.

When entitlement refuses, it throws an ExecutionRefusedError carrying the executionId and the port’s reasonCode, so a refusal cannot be mistaken for an empty result. An error from the run itself is recorded, the hold is released, and the error is rethrown unchanged — the caller still sees Mastra’s own.

Streaming

A stream’s usage is known only once it ends, so streamWithExecution returns the native stream at once, with the two ways a stream finishes:

import { streamWithExecution } from "@intelligo-dev/mastra";

const { stream, settle, abort } = await streamWithExecution(
  { executions, workspaceId, userId, capability: "support.reply" },
  () => supportAgent.stream(messages)
);

settle(result?) records usage when the stream finishes; abort(error?) is for a client disconnect or a timeout. Both are idempotent through the execution, so a route can wire every one of its finish, error and abort hooks without racing itself.

Entry points

• @intelligo-dev/mastra

npm · source