Capturing Workflows: From Idea to Published SOP

OpsBrain turns your team's know-how into structured, runnable workflows. Whatever you start from, a blank page, a plain-language description, a screen recording, or an existing document, the journey is the same: capture, review, refine, publish.

A few ground rules before you start:

• Only managers can create workflows. All four creation routes are manager-only. Workers and contributors will not see the creation options. Contributors can still edit an existing workflow's steps, but only managers can publish.
• Nothing goes live until you publish. Every capture method produces a draft you can review and edit first.
• Trial workspaces have a workflow allowance. A trial lasts 7 days and includes up to 2 workflows and 5 members. If your organization is over its allowance, opening a creation page redirects you back to the library with an upgrade prompt.

Click New workflow in the Workflows library to open the Capture Process menu. It offers four cards, and the same four cards also appear on the empty-state dashboard and in the guided tour. Press Esc or click outside the menu to close it.

The manual path. You fill in the basics, then build every step yourself in the editor.

1. 1Open New workflow and choose Create from Scratch. You land on a single card headed Basic Information. The page title reads Create New Workflow until you type a name.
2. 2Enter a Workflow Name. The placeholder suggests the shape: "e.g., Customer Onboarding Process". This field is required, and Continue stays disabled until it has a value. Names can be up to 255 characters.
3. 3Add a Description if you like. The helper text calls it optional but recommended for clarity. Descriptions can be up to 2,000 characters.
4. 4Pick Assigned Team(s) using the Select Team dropdown. Each team you pick appears as a removable pill; click the × on a pill to remove it. When every team is already picked, the list shows All teams added.
5. 5Check the Owner field. It is read-only and shows your name with "(You)". You are set as the guide owner by default.
6. 6Click Continue to create the workflow and open the editor, or Cancel to go back to the library.

Describe the process in plain language and OpsBrain builds a structured draft: steps, decision points, instructions, and documentation requirements. You review and edit everything before publishing.

1. 1Choose Generate with AI. The page is headed Generate SOP with AI.
2. 2Enter a Workflow Name (placeholder: "e.g., Warehouse Receiving Process").
3. 3Write your description in Describe the process. The placeholder tells you exactly what to cover: what the process does, who performs it, the steps in order, any decisions or branches, and anything that must be recorded or attached.
4. 4Watch the live character counter. It starts at "Minimum 60 characters", counts down how many more you need, and once you pass the floor switches to a running total out of 6,000.
5. 5Click Generate workflow. The label changes to Generating… while the request is submitted, then you move to the processing screen.

Limits and validation

• Minimum 60 characters, maximum 6,000. Both are enforced in the form and on the server.
• Gibberish detection. Once you pass 60 characters, OpsBrain checks that the text looks like a real description. If it does not, an inline warning appears: "This doesn't look like a real description. Write it in plain language so the AI can build a useful workflow."
• The disabled button explains itself. Hover the disabled Generate workflow button and a tooltip tells you what is missing: add a name, add at least 60 characters, trim to 6,000 or fewer, or rewrite gibberish in plain language.

Need a starting point?

Below the form, the panel Need a starting point? Try one of these: lists clickable example prompts. Clicking one fills the textarea and auto-derives the workflow name from the example's first five words. The auto-naming stops as soon as you edit the name yourself.

Walk through the process on your screen exactly as a teammate would. OpsBrain transcribes the audio and extracts structured steps with frames from the video.

1. 1Choose Record Guide. The page is headed Record your screen.
2. 2Enter a Workflow Name (placeholder: "e.g., New customer activation in CallUp").
3. 3Click Start recording. A setup dialog titled Before you start recording walks you through three things: microphone access (your browser asks to use your mic before the screen picker appears), choosing what to share (a window, a tab, or your full screen; recording starts the moment you confirm), and stopping when you are done (come back and click Stop, or use the browser's stop-sharing button).
4. 4Decide on audio. The checkbox Include microphone audio is recommended if you plan to narrate; your voice is mixed with any system audio. If you uncheck it, the dialog switches to a "No microphone" explanation: you can still capture system audio if your browser supports it.
5. 5Click Start recording in the dialog (it reads Requesting access… while permissions are granted). Mic permission is requested first, then the screen picker appears.
6. 6Do the work. The status line reads Ready to record before you start, then shows a running "Recording" timer in minutes and seconds while you capture.
7. 7Click Stop when finished. The status changes to a "Recorded" line with the clip length and an invitation to review and continue. An inline preview player lets you watch the clip back. Not happy with it? Click Record again.
8. 8Click Continue to upload the clip and start processing.

If something goes wrong

• "Screen recording is not supported in this browser. Use the latest Chrome, Edge, or Firefox on desktop." Mobile browsers cannot record.
• "Your browser refused to start the recorder. Try recording again, or use Chrome on desktop."
• "Recording was cancelled." appears if you dismiss the browser's share picker.
• "Record something before continuing." appears if you click Continue with no clip.

Already have the SOP written down or recorded? Upload it and OpsBrain will analyze the file and extract a draft workflow you can refine before publishing.

1. 1Choose Import Guide. The page is headed Import from a document or video.
2. 2Enter a Workflow Name. It is required.
3. 3Add your file. Drag it onto the dropzone that reads Drop a file or click to browse, or click Choose file. The dropzone lists what it takes: "PDF, MP4, MOV, WEBM, MP3, WAV, up to 20 MB".
4. 4Confirm the selection. The chosen file shows its name and size, with a Remove file button (×) if you picked the wrong one.
5. 5Submit. A progress line shows "Uploading" with a live percentage, then you move to the processing screen.

What you can import

If you submit without picking a file, the form reminds you: "Choose a file to import."

After you generate, record, or import, OpsBrain shows a live status page while the AI does its work. The page checks for updates every 3 seconds, so you can watch it move on its own. The icon tells you where things stand: a spinner while queued, pulsing sparkles while processing, a green check on success, a red triangle on failure. A chip shows the file name and its current status.

On success the title becomes Workflow draft is ready (or Updated draft is ready after a refinement) and the page redirects automatically after about a second and a half, either to the editor via Open draft, or to the structure review screen via Review structure when OpsBrain found multiple procedures in the document.

On failure the title reads We could not finish the import. Something went wrong while extracting the workflow; you can try again, or click Build manually to start from what was recovered. The raw error appears in a red box so you can share it with support. Back to library is always available in the header.

Some paths end on a Workflow Created Successfully screen with three ways forward: View Workflow, Start Run, and Create Another Workflow.

The editor (headed Edit guide) is where drafts become polished SOPs. The layout has three parts: an outline rail on the left listing every step, the step cards in the middle, and a Step configuration inspector on the right. Until you select a step, the inspector shows Configure a step and asks you to pick one from the list. An empty workflow greets you with No steps yet. Add your first step.

8.1 Adding steps

Hover between steps (or below the last one) to reveal a dashed divider with a + Add step pill. Clicking it opens a picker headed Choose a step type with eight types:

8.2 Settings every step shares

• Step title: the short imperative line the worker sees first (placeholder "Enter step title").
• Instructions: the how-to detail (placeholder "Enter instructions for this step"). Long Text and Sub-Checklist steps get a taller box.
• Required step toggle, off by default. Required steps cannot be skipped without a written reason.
• Allow worker notes toggle, on by default. Controls whether workers can leave a note on this step during a run.
• Row actions: Duplicate step, Delete step, and a drag handle (Drag to reorder).

8.3 Two advisories worth reading

The editor nudges you when you flip either default, and the advice is deliberate:

8.4 Type-specific editors

• Decision: an Options panel lists each answer (Option 1, Option 2, and so on) with an Add option button and per-option remove buttons. New Decision steps start with Yes and No. Renaming an option keeps its branch routing intact.
• Legacy yes/no Decision: a single Decision question field ("e.g., Is all required information present?") with fixed Yes/No branching.
• Sub-Checklist: a Sub-checklist panel with Add checklist item. The empty state reminds you to click it to create the first item. A sub-checklist must have at least one item before the workflow can publish.

Branching is how one workflow adapts to reality. Only Decision steps (and legacy yes/no steps) branch.

9.1 The Branching logic panel

On a Decision step, the Branching logic panel shows one row per answer: if the answer is X, go to a destination you pick. The footer reminds you that the workflow automatically continues based on the worker's answer. Each destination dropdown offers three kinds of target:

• Another step in this workflow, listed as "Step n: title". The natural next step is marked with "· next".
• Launch another workflow, listed as ↗ Launch "Name". Unpublished targets are marked "(Draft)".
• End workflow, listed last, which finishes the run on that answer.

By default, new branches point at the next sequential step, so a fresh Decision records the answer without changing the path until you route it.

9.2 Launching sub-workflows: the three after-launch modes

When a branch launches another workflow, an After the launched workflow choice appears with three mutually exclusive modes:

Below the mode picker, a branch target preview shows the sub-workflow's first step ("opens on…"), which team owns it, and a cross-team — pending pickup marker when the target belongs to a different team. Cross-team launches land in that team's queue for someone there to claim; the parent run waits or hands off depending on the mode.

9.3 The branch warnings badge

The editor keeps an eye on "dead decisions". When Decision steps send every answer to the same next step, a badge titled Decisions worth a look appears with a count to review. That pattern is fine if you only want to record the answer, but if an answer should change the path, open the step and set where each answer goes.

10.1 Reference attachments

Each step has a Reference attachments (optional) panel for uploading reference images, PDFs, or documents workers should review. Click Upload file to add one. Files up to 20 MB are accepted, any type.

• On a step that has not been saved yet, uploads wait until the workflow saves; the panel notes that the step will be saved automatically before uploading.
• Removing an attachment asks Remove this attachment? and names the file. Removed attachments are kept recoverable behind the scenes and can be restored.
• Success and failure are confirmed with toasts: "Attachment uploaded.", "Attachment removed.", or "Couldn't upload the file."

10.2 Autosave and undo

• Autosave runs continuously: a "Saving…" indicator settles into Auto-saved. If a save fails, a toast says "Couldn't autosave, your edits are still on screen." so you can retry without losing work.
• Undo and Redo live in the header, with keyboard shortcuts ⌘+Z and ⌘+Shift+Z. The buttons name what they will undo, for example "Undo: Reorder steps". History covers reordering, adding, deleting, duplicating, and edits.
• The outline rail marks steps that branch with Has branching and problem steps with Needs attention.

10.3 The flow diagram

Click View flow diagram in the header to see the whole workflow as a flow chart. Decision steps fan out, sequential steps flow down, and the legend distinguishes sequential links, branch jumps, branches that launch workflows, and end points. Hover any node to highlight its connections, and click any launched workflow to jump into it.

Rather than hand-editing dozens of steps, you can ask OpsBrain to rework the draft. Click Refine with AI in the editor header. The modal has two tabs, and both produce a fresh draft that you review before publishing. Nothing changes for workers until you publish.

Mode 1: Describe a change

Describe the change in plain language. OpsBrain keeps the steps you do not mention and produces a revised draft.

• Four one-click presets cover the common asks: Make it more detailed, Make it shorter and simpler, Add evidence/photo steps where useful, and Add a final approval step at the end.
• Or type your own instruction, up to 2,000 characters. The placeholder shows the level of specificity that works well: "e.g., Split receiving and put-away into two separate stages, and add a step to photograph any damaged goods."
• Instructions need at least 3 characters, and gibberish is rejected with "That doesn't look like a real instruction. Describe the change in plain words."
• Click Regenerate draft to submit.

Mode 2: Replace from a file

Upload a newer PDF or recording of the process. OpsBrain re-reads it and rebuilds the workflow into a fresh draft.

• Same formats and limit as Import: PDF, MP4, MOV, WEBM, M4A, MP3, WAV, up to 20 MB.
• Submitting without a file prompts: "Choose a document or recording to re-read."
• Click Reprocess workflow to submit.

Both modes go through the processing screen (with "Regenerating your workflow" copy, usually 20 to 60 seconds) and land as a draft for your review.

Publishing makes a draft the live version that workers run. Editing a live workflow never mutates it in place: OpsBrain quietly forks a new draft, so runs already in progress keep their exact steps until they finish. Only managers can publish or revert.

12.1 Validation: what blocks a publish

OpsBrain refuses to publish an unrunnable workflow. You will see one of these exact messages:

• "A workflow must have at least one step before it can be published."
• "Step "…" (sub-checklist) must have at least one checklist item."
• "Step "…" (multiple choice) must have at least one option."
• "Step "…" launches "…" on "…", but that workflow hasn't been published yet. Publish it first, then publish this one."

12.2 Publishing sub-workflows together

If your branching launches workflows that are still drafts, publishing opens a modal titled Publish these sub-workflows too? (eyebrow: Almost there). It lists every draft child, each badged Currently a draft, and explains that publishing them now lets workers run the full routine end to end. If those children launch further drafts of their own, the whole chain goes live together, in one transaction, deepest workflows first. Confirm with Publish all & go live.

12.3 What publishing does

• Flips the workflow's status to Live and makes this version the one new runs use.
• Leaves in-flight runs untouched; they finish on the version they started on.
• Marks pending optimization suggestions that were based on older versions as Stale.
• Emits a workflow.published event for your connected integrations and refreshes trigger field-mapping health.

12.4 Version history and revert

The workflow's Versions tab is headed Version history. Every publish is preserved. You can search change notes, and filter by All, Edits, Reverts, Drafts, or Current; each version carries a kind badge (Initial, Edit, Revert, or Draft). Change notes are written automatically by the system, for example "AI-extracted from uploaded SOP document" or "Reverted to v2".

To roll back, click Revert to this on any version. The confirm modal, Revert to v…?, explains the model: OpsBrain does not rewind. It duplicates that version's content into a brand-new version and publishes it immediately, so the old version stays untouched and your audit trail stays linear (v1, v2, v3 as a revert of v1). Active runs finish on their current version; new runs use the freshly published one. Confirm with Publish v… content, and you will see "Reverted to v…, a new version was published."