@ingram-cloud/flue

Drive an Ingram Cloud smith from a Flue agent. A thin, idiomatic extension of Flue: register a smith as a model provider, attach Ingram-hosted tools over MCP, and resolve approvals — each over an industry-standard surface, no bespoke protocol.

Philosophy: stand on the standard

This package adds no proprietary client. The provider rides Flue's built-in openai-completions wire protocol pointed at Ingram Cloud's OpenAI-compatible API; tools ride MCP; approvals ride the standard tool-call channel. A smith still runs the agent loop server-side — memory, tools, approvals, isolation — but to Flue it looks like any model.

Install

npm install @ingram-cloud/flue

@flue/runtime (v1 beta) is a peer dependency — you already have it in a Flue app. It must stay a peer: registerProvider() writes to a module-scoped registry, so the adapter and your app have to share one instance.

Provider: a smith as your model

Register once in src/app.ts, before routing — exactly like any other Flue provider:

import { registerProvider } from "@flue/runtime"; // (used by the adapter)
import { registerIngramCloud } from "@ingram-cloud/flue";
import { flue } from "@flue/runtime/routing";
import { Hono } from "hono";

// A per-smith token names exactly one smith; the agent is the one that smith runs.
export const ingram = registerIngramCloud({ apiKey: process.env.IC_SMITH_TOKEN! });

const app = new Hono();
app.route("/", flue());
export default app;

Then point an agent at it:

// agents/triage.ts
import { defineAgent } from "@flue/runtime";
import { ingram } from "../src/app.js";

export default defineAgent(() => ({
  model: ingram.model("gpt-5.5"),
  instructions: "Triage the incoming request.",
}));

The model id is the LLM, not the agent

The agent a smith runs — its instructions, tools, and memory — is resolved from the smith, never from the model id. The model id is the upstream inference LLM for the turn (ingram.model("gpt-5.5") runs the smith's agent on GPT-5.5).

Unlike the raw OpenAI-compatible surface, Flue requires a non-empty model id (provider/model), so there is no "use the agent's configured model" form here — name the model you want. ingram.model("") throws to make that explicit.

Server-side with a tenant-admin token instead of a smith token? Name the smith:

const ingram = registerIngramCloud({
  apiKey: process.env.IC_TENANT_TOKEN!,
  smithId: "smt_…",
});

Never ship a tenant-admin token to the browser.

Tools: Ingram-hosted MCP

Expose an Ingram Cloud deployment of kind: "mcp" as Flue tools. Spread them into defineAgent({ tools }):

import { connectIngramMcp } from "@ingram-cloud/flue";

const ingramTools = await connectIngramMcp({
  apiKey: process.env.IC_SMITH_TOKEN!,
  deploymentId: "dep_…",
});

export default defineAgent(() => ({
  model: ingram.model("gpt-5.5"),
  tools: [...ingramTools.tools], // named mcp__ingram__<tool>
  instructions,
}));

Tools arrive as ordinary Flue tools, so they compose with native tools, skills, and the sandbox. Call ingramTools.close() when they're no longer needed.

Approvals (human-in-the-loop)

A smith tool marked destructiveHint pauses the run for approval. On this surface the pause arrives as a tool call whose id is "<run_id>::<tool_call_id>" and the turn ends with finish_reason: "tool_calls". The helpers are pure and also live at @ingram-cloud/flue/approvals:

import {
  getApprovalRequests,
  approvalWireMessage,
} from "@ingram-cloud/flue";

// `toolCalls` from the model response (composite ids are Ingram approvals).
const approvals = getApprovalRequests(toolCalls);
for (const a of approvals) {
  const decision = (await askTheHuman(a)) ? "approve" : "reject";
  // Send this `tool` message back as the next turn to resume the paused run:
  sendNextTurn(approvalWireMessage(a, decision));
}

On approve, Ingram Cloud executes the tool itself and continues; on reject, the run completes with stop_reason: "approval_rejected" and nothing runs.

Identity & tokens

| Token | Use | How the smith is chosen | |---|---|---| | Smith token (sub = "<tenant>:<smith>") | browser-safe; the default | the token is the smith | | Tenant-admin token | server-side only | pass smithId (sent as IC-Smith-Id) |

Notes

• ESM-only, ships as dist/. Build with npm run build (plain tsc).
• Independent of the API's api/web checks, like the pulumi/ and ai-sdk/ packages. Keep it in step when the OpenAI-compatible / MCP surfaces it wraps change.
• For the same product over the Vercel AI SDK, see @ingram-cloud/ai-sdk.