Workflow products

A workflow product lets you buy a finished job shape instead of choosing a model. Rather than naming gpt-4o-mini and a routing mode yourself, you pick a curated outcome — classify, extract, summarize, and so on — and BatchRouter selects the model and provider lanes that satisfy that contract. You describe the job; BatchRouter handles the infrastructure choices.

This is the outcome-contract alternative to the model-quote path. Both end at the same place — a quote_id you pass to POST /v1/batches — but they differ in who picks the model.

When to use a workflow product

A workflow product is a curated, versioned outcome contract plus a preset manifest (input shape, output shape, SLA, route policy). It is not a general DAG, pipeline, or multi-step orchestration engine — it is a single curated job outcome that BatchRouter knows how to route well.

Reach for a workflow product when:

• You care about the result shape (a clean classification, a structured extraction, a summary) more than which model produces it.
• You want BatchRouter to balance fit, validation history, and cost rather than always taking the cheapest model.
• You’d rather not track model slugs, capabilities, or provider lanes yourself.

Use the model-quote path instead when you need a specific model, an explicit fallback list, or fine control over routing_mode.

Note

Either path produces a quote_id. Quotes are always free — credits are only reserved when you create the batch.

Browse the catalog

List the available workflow products to discover slugs and their default SLA tier.

cURL

curl https://api.batchrouter.com/v1/catalog/workflow-products \
  -H "Authorization: Bearer $BATCHROUTER_API_KEY"

Node

const res = await fetch("https://api.batchrouter.com/v1/catalog/workflow-products", {
  headers: { Authorization: `Bearer ${process.env.BATCHROUTER_API_KEY}` },
});
const { data } = await res.json();

Python

import os, requests

res = requests.get(
    "https://api.batchrouter.com/v1/catalog/workflow-products",
    headers={"Authorization": f"Bearer {os.environ['BATCHROUTER_API_KEY']}"},
)
data = res.json()["data"]

Each entry carries a slug, display_name, description, sla_tier, and a latest_version_id. Fetch a single product by slug to inspect its details before quoting:

cURL

curl https://api.batchrouter.com/v1/catalog/workflow-products/classify \
  -H "Authorization: Bearer $BATCHROUTER_API_KEY"

Node

const res = await fetch(
  "https://api.batchrouter.com/v1/catalog/workflow-products/classify",
  { headers: { Authorization: `Bearer ${process.env.BATCHROUTER_API_KEY}` } },
);
const { product } = await res.json();

Python

res = requests.get(
    "https://api.batchrouter.com/v1/catalog/workflow-products/classify",
    headers={"Authorization": f"Bearer {os.environ['BATCHROUTER_API_KEY']}"},
)
product = res.json()["product"]

Tip

classify is illustrative. List the catalog to see the real slugs available to your account, and check the interactive API reference for the full field list on each product.

Quote a workflow product

POST /v1/quotes/workflow quotes a curated workflow product and returns model-agnostic provider lanes that satisfy the contract. Send the workflow slug plus your input(s). You can quote a single item with input, or a representative set with inputs (each entry takes a customer_item_id and an input object).

cURL

curl https://api.batchrouter.com/v1/quotes/workflow \
  -H "Authorization: Bearer $BATCHROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "workflow": "classify",
    "inputs": [
      {
        "customer_item_id": "item-1",
        "input": {
          "messages": [
            { "role": "user", "content": "Summarize: BatchRouter routes batch-AI workloads across providers." }
          ]
        }
      }
    ]
  }'

Node

const res = await fetch("https://api.batchrouter.com/v1/quotes/workflow", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.BATCHROUTER_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    workflow: "classify",
    inputs: [
      {
        customer_item_id: "item-1",
        input: {
          messages: [
            { role: "user", content: "Summarize: BatchRouter routes batch-AI workloads across providers." },
          ],
        },
      },
    ],
  }),
});
const { quote_id, pricing_estimate, quote_lanes } = await res.json();

Python

res = requests.post(
    "https://api.batchrouter.com/v1/quotes/workflow",
    headers={
        "Authorization": f"Bearer {os.environ['BATCHROUTER_API_KEY']}",
        "Content-Type": "application/json",
    },
    json={
        "workflow": "classify",
        "inputs": [
            {
                "customer_item_id": "item-1",
                "input": {
                    "messages": [
                        {"role": "user", "content": "Summarize: BatchRouter routes batch-AI workloads across providers."}
                    ]
                },
            }
        ],
    },
)
quote = res.json()
quote_id = quote["quote_id"]

The response is the same QuoteResponse shape as a model quote: a quote_id, a pricing_estimate, and the quote_lanes BatchRouter would route to. The difference is that you never named a model — BatchRouter chose lanes that satisfy the workflow’s outcome contract.

Pin a workflow version

Workflow products are versioned. Passing the bare slug uses the latest version. To pin a specific version for reproducibility, send workflow as an object with slug plus either version_id or version_number:

{
  "workflow": { "slug": "classify", "version_number": 3 },
  "input": { "messages": [{ "role": "user", "content": "…" }] }
}

Constrain cost and tools

• max_price caps the quote — see the API reference for the Money shape.
• required_tools forces provider-hosted tools (such as web_search or python_execution) onto every lane. Quote-level required_tools are merged with any per-input requirements, and lanes whose provider doesn’t declare every required tool are rejected with a tool_support failure. If no eligible lane remains, the quote fails rather than routing to a lane that lacks the tools.

How it differs from a model quote

The two quote endpoints converge on a quote_id but invert who owns model selection.

	Model quote (POST /v1/quotes/model)	Workflow quote (POST /v1/quotes/workflow)
You choose	model / models, routing_mode	a workflow slug (and optional version)
BatchRouter chooses	the cheapest eligible lane for your model(s)	the model + provider mix that satisfies the contract
Input	items[] (each names a model)	input or inputs[] (model-agnostic)
Best when	you need a specific model or routing mode	you care about the outcome, not the model
Returns	quote_id, pricing_estimate, quote_lanes	quote_id, pricing_estimate, quote_lanes

Caution

For workflow products the cheapest lane does not always win — the router balances workflow fit and validation history against cost, so a workflow quote may price differently than the cheapest model that could nominally do the job.

From quote to batch

Once you have a quote_id, the rest of the flow is identical to any other batch.

1. Create the batch — pass the quote_id to POST /v1/batches with an Idempotency-Key header. The active quote snapshot reserves credits; final cost settles from actual provider token usage.

2. Poll — GET /v1/batches/{batchId} until a terminal status (completed, or failed | cancelled | expired).

3. Fetch results — page GET /v1/batches/{batchId}/results, or pull the signed bundle from GET /v1/batches/{batchId}/artifact-url.

See Submit a batch for the full create-poll-results walkthrough.

Next steps

Quotes and pricing The model-quote path and how estimates become reserved credits.

Models and catalog Browse models and providers when you want to pick the model yourself.

Submit a batch Turn any quote into a running batch and retrieve results.

API reference Full request and response fields for every endpoint.