Recipes

A recipe is a proven combination of ModuleX features that solves one real problem. Each one below names the pieces it stitches together — the Assistant, the AI Composer, a workflow, your knowledge, and your integrations — tells you when to reach for it, and points to the guide or example that walks it end to end. Use this page as a menu. Skim the cards, find the pattern closest to what you are trying to do, and follow the links into the step-by-step guides. For a single Assistant task carried out from prompt to result, see Assistant examples. For a workflow run start to finish, see run a workflow.

Pick a recipe

Ask-and-act with the Assistant

One plain-language request, several connected tools, a result. Best for one-off tasks across your services.

Prompt to running workflow

Describe an automation in English, let the Composer build it on the canvas, then run it on demand or on a schedule.

Answer from your own documents

Ground the Assistant or a workflow in your knowledge bases so answers come from your content, not the open web.

Scheduled automation

Run a workflow automatically on a cron schedule, or trigger it from your own code over the API.

Approve before acting

Let an agent or workflow gather context and draft, then pause for a human to approve the one step that changes something.

Co-build with your team

Two or more people edit the same workflow on one canvas in realtime, with the Composer helping alongside.

How to read each recipe

Every recipe follows the same shape, so once you have read one you can skim the rest.

What it combines

The ModuleX features the recipe stitches together — the parts you already know from the feature map.

When to use it

The kind of problem this pattern is the right fit for, and when a different recipe would serve you better.

Where to go next

The guide, example, or concept page that walks the pattern end to end.

Ask-and-act with the Assistant

The simplest recipe. You type one request in plain language and the Assistant does it — reading from your connected services, deciding the next step, and calling one tool at a time until the task is done. What it combines: the Assistant and your connected integrations. No workflow, no canvas, no setup beyond connecting the services you want it to use. When to use it: a one-off task that does not need to run again. “Summarize the newest issues in this repo,” “draft a reply to the latest support email,” “pull these numbers and tell me what stands out.” If you find yourself asking for the same thing repeatedly, graduate to the next recipe and turn it into a workflow.

Connect the services it needs

The Assistant can only use tools you have connected. If a request needs a service you have not added, it asks you to connect it right inside the chat — see connect an integration.

Ask in plain language

Send one message that names what to find and what to do with it. One message starts one turn.

Watch it work

The Assistant streams its progress live: which tool it is calling, what it found, and what it plans next. If a step would change or send something, it pauses for your approval first (see approve before acting).

Where to go next

Four Assistant tasks worked end to end — research and summarize, inbox triage, draft and send, and a multi-tool task.

Prompt to a running workflow

When a task is worth repeating, turn it into a workflow. The fastest way to build one is to describe it in plain English and let the AI Composer generate an editable workflow graph on the canvas — then you refine it and run it as often as you like. What it combines: the AI Composer to build, the workflow builder canvas to refine, and integrations or LLM nodes to do the work. When to use it: any task you will run more than once, or that has more than a couple of steps, or that you want to trigger on a schedule or from code later. Unlike a one-off Assistant request, a workflow is saved, versioned, and reusable.

Describe the automation

Tell the Composer what you want in plain language. It turns your description into a complete, editable workflow graph — nodes, connections, and all.

Refine on the canvas

Adjust the generated graph: swap a model, add a tool node, branch with a conditional, or keep chatting with the Composer to make changes for you.

Run it

Run the workflow from the builder and watch the live result, or call it from chat, the API, or an SDK. See run a workflow.

Where to go next

Build a workflow with Composer — from a plain-English prompt to a running workflow, step by step.

Answer from your own documents

Make the Assistant or a workflow answer from your content — policies, product docs, past tickets — instead of guessing. ModuleX retrieves the relevant passages from your knowledge bases and grounds the answer in them. What it combines: a knowledge base (managed or your own vector store) plus either the Assistant or a knowledge node inside a workflow. When to use it: any question where the right answer lives in your own documents. Support replies that must match your policies, internal Q&A over a handbook, or a workflow step that needs company context before it drafts. Reach for this whenever a generic model answer is not good enough because it does not know your specifics.

Build a knowledge base

Create a knowledge base and upload your documents. ModuleX parses, chunks, and indexes them for retrieval — see build a RAG knowledge base.

Point a question at it

Ask the Assistant something like “Summarize what our onboarding guide says about SSO,” or drop a knowledge node into a workflow to retrieve context for a later step.

Get a grounded answer

The answer comes from your documents, with the retrieved context behind it. For chat-style use, see chat with your knowledge.

Searching managed knowledge (the ModuleX-hosted vector store) uses a small amount of credit per retrieval. Bringing your own vector store is not metered in credits. See credits and metering.

Where to go next

Create a knowledge base, ingest documents, and query it.

Scheduled automation

Once a workflow does the right thing, you rarely want to press Run by hand. This recipe makes it happen on its own — on a clock, or triggered by your own systems. What it combines: a saved workflow plus either a schedule or a call to the run API. When to use it: recurring jobs (a daily digest, a weekly report, an overnight sync) belong on a schedule. Event-driven jobs (run this when an order is placed, when a form is submitted) belong on an API trigger from the system that knows about the event.

On a schedule
From your own code

Attach a cron schedule to a workflow and ModuleX runs it for you. A daily summary, a Monday-morning report, an hourly check — all set up once and left alone.Schedules and other automated runs require an organization owner or admin role. See organizations, roles & membership.

Where to go next

Schedule a workflow to run automatically on a cron schedule.

Trigger a workflow from anywhere with one authenticated call. Every request carries Authorization: Bearer mx_live_… plus your X-Organization-ID. The call returns a run_id you can then stream over SSE.

curl -X POST https://api.modulex.dev/workflows/run \
  -H "Authorization: Bearer mx_live_xxx" \
  -H "X-Organization-ID: 11111111-1111-1111-1111-111111111111" \
  -H "Content-Type: application/json" \
  -d '{"workflow_id": "wf_123", "input": {"topic": "weekly digest"}}'
# → {"status": "running", "run_id": "6a7b...", "thread_id": "550e..."}

from modulex import Modulex

client = Modulex(
    api_key="mx_live_xxx",
    organization_id="11111111-1111-1111-1111-111111111111",
)

run = await client.executions.run(
    workflow_id="wf_123",
    input={"topic": "weekly digest"},
)
print(run.run_id)

import { Modulex } from "modulex-js";

const client = new Modulex({
  apiKey: "mx_live_xxx",
  organizationId: "11111111-1111-1111-1111-111111111111",
});

const run = await client.executions.run({
  workflowId: "wf_123",
  input: { topic: "weekly digest" },
});
console.log(run.runId);

A run on a charged surface can be turned away by the billing gate with a 402, 403, or 429 response before any work starts — handle that in your trigger. See run a workflow for the full walkthrough.

Approve before acting

The safest way to let automation touch the real world: have it gather context and prepare the change, then stop and wait for a person to say yes. ModuleX calls this human-in-the-loop. What it combines: an approval pause — the Assistant stopping before a sensitive action, or an interrupt node inside a workflow — plus whatever tools do the reading and the writing. When to use it: any time a step sends a message, creates or deletes a record, charges money, or otherwise changes something you would not want done without a look first. Combine it with ask-and-act for one-off tasks, or with a workflow for repeatable ones.

Let it prepare

The agent or workflow reads what it needs and drafts the change — a reply, a ticket, an update — without committing it yet.

Pause for a person

Before the step that writes, it pauses and shows exactly what it is about to do on an approval card. Only the person who started the turn can answer.

Approve, edit, or decline

Approve to let it proceed, reply with a change to have it revise and ask again, or decline to stop. Nothing risky happens until you approve.

The Assistant always pauses before a sensitive action — a draft is never sent and a record is never created until you approve it. In a workflow, you add this pause yourself with an interrupt node. See human-in-the-loop for the question and approval kinds.

Where to go next

See the “draft and send” example for the canonical pause-before-acting pattern.

Co-build with your team

Workflows are not a solo activity. Several people can edit the same canvas at the same time, see each other’s cursors and selections, and let the Composer make changes alongside everyone — all in realtime. What it combines: realtime co-editing on the canvas plus the AI Composer editing the same workflow. When to use it: building or reviewing a workflow as a team, walking a teammate through an automation, or pairing with the Composer while a colleague refines another part of the graph. Edits sync live, with presence and node locks so two people do not fight over the same node.

Open the same workflow

Invite a teammate and open the workflow on the canvas. You both see who is present and where they are working.

Edit together

Changes appear for everyone as they happen. Node locks keep two people from editing the exact same node at once.

Let the Composer help

Ask the Composer to make a change and it edits the live workflow; the update syncs to every collaborator on the canvas. This works over the Socket.io collaboration channel — see realtime & collaboration model.

Where to go next

Invite a teammate and co-edit a workflow live.

Combine recipes

The best results come from chaining recipes. A few patterns worth knowing:

Grounded, scheduled digest

Build a workflow with the Composer, give it a knowledge node so it pulls from your own documents, then put it on a schedule. Result: a recurring digest written from your content, with no one pressing Run.

Triage then act, with a checkpoint

Start with ask-and-act to sort an inbox, then ask the Assistant to take action on the top items. The approve-before-acting pause kicks in automatically the moment a step would change something.

One-off proven, then promoted

Solve a problem once with the Assistant. When you find yourself asking for the same thing again, hand the prompt to the Composer to turn it into a reusable, schedulable workflow.

Keep going

Feature map

Every ModuleX feature, what it does, and where to use it — the building blocks behind these recipes.

Optimization

Tune your workflows and usage for speed, reliability, and cost.

Assistant examples

Four Assistant tasks worked end to end, ready to copy and adapt.

Run a workflow

Authenticate, run a workflow, and stream the result in three languages.

Billing & credits

Data & limits

Guides

Power using

Pick a recipe

Ask-and-act with the Assistant

Prompt to running workflow

Answer from your own documents

Scheduled automation

Approve before acting

Co-build with your team

How to read each recipe

What it combines

When to use it

Where to go next

Ask-and-act with the Assistant

Where to go next

Prompt to a running workflow

Where to go next

Answer from your own documents

Where to go next

Scheduled automation

Where to go next

Approve before acting

Where to go next

Co-build with your team

Where to go next

Combine recipes

Keep going

Feature map

Optimization

Assistant examples

Run a workflow

​Pick a recipe

Ask-and-act with the Assistant

Prompt to running workflow

Answer from your own documents

Scheduled automation

Approve before acting

Co-build with your team

​How to read each recipe

What it combines

When to use it

Where to go next

​Ask-and-act with the Assistant

Where to go next

​Prompt to a running workflow

Where to go next

​Answer from your own documents

Where to go next

​Scheduled automation

Where to go next

​Approve before acting

Where to go next

​Co-build with your team

Where to go next

​Combine recipes

​Keep going

Feature map

Optimization

Assistant examples

Run a workflow

Pick a recipe

How to read each recipe

Ask-and-act with the Assistant

Prompt to a running workflow

Answer from your own documents

Scheduled automation

Approve before acting

Co-build with your team

Combine recipes

Keep going