How-to guides
Building Automated Workflows
Turn a repeatable task into a saved, runnable workflow.
A workflow is a saved sequence of steps OOMU can run for you; draft a reply, summarise a folder, call a connected tool, wait for your approval, produce a file. You describe what you want in plain language, OOMU composes the workflow for you, and you can run it on demand or on a schedule.
This guide covers building, refining, running, and scheduling a workflow.
How workflows are built
You don't wire workflows together by hand. Instead, OOMU composes them:
- You write a plain-language description of what the workflow should do.
- OOMU's on-device model drafts the workflow against a capability catalogue; the set of things it's actually able to do right now.
- You review the draft, refine it in natural language, and save it.
Because OOMU only builds from capabilities that genuinely exist in your setup, the workflows it produces are runnable, not aspirational.
The capability catalogue
Every workflow step comes from a capability. There are four kinds:
| Capability kind | What it does |
|---|---|
| Agent | Runs one of your assistants; draft, summarise, classify, decide. |
| Control | Structural logic; route between branches, apply a condition, loop, or gate on a permission. |
| MCP tool | Calls a tool exposed by a connected MCP server (see below). |
| System action | Performs a defined local action in your workspace. |
Each capability is marked available or requires connection. A capability that needs an external tool shows as *requires connection* until you've connected the MCP server that provides it, so you always know what a workflow depends on before you run it.
MCP (Model Context Protocol) servers are how OOMU connects to external tools and data sources. When a server is connected, its tools appear in the catalogue and can be used as workflow steps. If a workflow needs a tool you haven't connected, OOMU flags it as a missing capability rather than failing silently.
Where workflows live. Workflows are a section of the Tasks door; open Tasks → Workflows for the composer and your saved library. Every time a workflow runs, that run appears in Tasks → Now with its progress and evidence, and a workflow put on a schedule becomes a Routine under Tasks → Scheduled.
Step 1; Describe the workflow
Open Tasks → Workflows and start a new workflow with the composer. Describe the outcome you want in plain language, for example:
"Read every document in my Contracts folder, summarise each into three bullet points, flag any that mention auto-renewal, and save the result as a single report I can open."
Be specific about inputs (a folder, a topic), the work (summarise, compare, extract), and the output (a file, an on-screen answer).
Step 2; Review the composed workflow
OOMU drafts the workflow as a sequence of connected steps and shows it to you as a storyboard; a readable, left-to-right view of what will happen, in order. Under the hood each step is a typed node (an input, an agent step, a router, a permission gate, a tool call, an output), but you read it as a plain sequence.
Check three things:
- Does the shape match your intent? Inputs at the start, your work in the middle, an output at the end.
- Are any steps marked as needing a connection? If so, connect the relevant MCP server, or ask OOMU to compose an alternative that doesn't need it.
- Are there permission gates where you'd want them? Anything that acts on your files or reaches outside should pause for approval (see Step 4).
Step 3; Refine in natural language
You don't have to get the description perfect the first time. Ask OOMU to edit the workflow conversationally:
"Add a step that emails me the report when it's done." "Only include contracts modified this year." "Put an approval checkpoint before anything is saved."
OOMU re-composes the affected parts against the catalogue and updates the storyboard. When you're happy, save it. Saved workflows are versioned, so you can keep refining without losing prior versions.
Step 4; Run it, with approval where it matters
Run a saved workflow on demand. As it executes, OOMU streams its progress step by step.
Workflows can include permission gates; checkpoints where execution pauses and waits for your explicit approval before doing something consequential (writing a file, calling an external tool, sending something out). When a workflow reaches a gate, you'll see exactly what it wants to do and can approve or decline. Nothing consequential happens behind your back.
When the workflow produces a file, OOMU gives you a direct way to open the output from the run.
Step 5; (Optional) Put it on a schedule
A workflow you run often can run itself. Turn it into a Routine (under Tasks → Scheduled) and OOMU will execute it automatically at the times you set; a Monday-morning summary, a nightly digest, a periodic sync, and can even message you the result. Scheduled runs honour the same approval gates as manual runs, so automation never means losing oversight.
Larger jobs: parallel helpers
For research-heavy work, a workflow's agent step can fan out to helpers; up to eight read-only workstreams that check different angles at once. Helpers *gather* information in isolated contexts; they can't write, send, or change anything. OOMU stays in charge: it synthesises their findings and keeps every consequential action for itself, behind your approval. You watch their progress, and can stop any one of them, under Tasks.
Good workflows to start with
- Summarise a folder into a single report.
- Draft grounded replies from a source document you provide.
- Extract and flag; pull specific facts from a batch of files and highlight the ones that need attention.
- Tool-assisted routines; combine a connected MCP tool with an agent step (e.g. fetch data, then summarise it).
Availability
Workflow authoring (composing and editing in natural language) is enabled by default in current builds. If your organisation has disabled it via configuration, the composer will tell you it's unavailable; saved workflows can still be run.
Related
- Scheduling Routines: turn a workflow into recurring, delivered work.
- Connecting Services: connect the tools a workflow step calls.
- Smart Model Routing: how the assistant steps inside a workflow choose local vs. cloud.
- The Delegated-Work Loop: how a run reports progress and evidence.
- Local Database Schema: how workflow definitions, runs, approvals, and schedules are stored.