One reference pattern
One reference pattern for supervised AI inside a mature product
Guidely is our own build of the guidance layer we engineer into vertical SaaS products. It is opened on a real record by a real role, offers the next useful move, then hand-holds the workflow through your own APIs and approval rules. In your product it ships under your name, in your design system, on your APIs and inside your permission model.
Form factors
Dialogue panel
Guided work, onboarding, occasional users
Wand palette
Power users, speed, keyboard-driven work
Animated character
Training-heavy workflows, low tech confidence
Inline nudge
High-frequency screens where a panel would be noise
A reference build, not software you licence. It ships under your name.
How it feels
Context first. Then the next useful move. Then the work.
This reference build already knows the account, module, record and role it was opened on. It says so, offers what is worth doing here, then hand-holds the workflow and stops for a human wherever your rules say it should. Switch between the shapes and personas to see the form factors we can build.
Guidely is our own build of this pattern, used here to show the idea. In your product it ships under your name, in your design system, on your APIs and inside your permission model.
Search bar. The universal search bar becomes the ask bar. One wand icon, same shortcut.
app.meridiancompliance.com / inspections / queue
Inspection queue
5 of 5 records
One new pixel: the wand in the top bar. Click it to ask anything.
Discovery
Users see what your product can do for them at this exact point in the workflow, instead of hunting through menus or waiting for a training call.
Hand-holding
Guidely walks the user through the workflow step by step, filling what it can infer and asking only for the decisions a person should own.
Its own run log
Context, steps taken, approvals requested and actions written all live in Guidely's own rail, tied back to your audit trail. Nothing is bolted into your product chrome.
Shapes it can take
A dialogue, a wand, a character on the page, or nothing at all.
All of them run on the same agent harness, so the context, permissions, approval rules and run log are identical. The choice is about who you are serving on that screen.
Dialogue panel
A docked conversation with the context header on top and the run log beside it. Talk or type.
Guided work, onboarding, occasional users
Wand palette
A keyboard-first command palette that returns navigation, actions and answers in one list. Talk or type.
Power users, speed, keyboard-driven work
Animated character
A character on the page that points at what it is talking about and offers the next move in a bubble. Talk or type.
Training-heavy workflows, low tech confidence
Inline nudge
No panel at all. One suggested next move rendered on the record itself, with the same harness behind it.
High-frequency screens where a panel would be noise
Talk to it or type at it, in every shape
Each surface carries a microphone next to the text field. A field engineer holding a tablet can speak the request, a manager at a desk can type it, and switching mid-task keeps the same context, the same permissions and the same run log. Voice-led personas read the answer back.
Useful for hands-busy roles such as field, warehouse, clinical and driving work, and for users who would rather ask than learn the interface.
Your name, your character, your product
Guidely is how we demonstrate the idea. In your product it ships under whatever name and character fits your customers. Jarvis when it should listen and speak. Copilot when it should feel like a teammate. Something quieter when it should stay out of the way. The persona is presentation. The context, permissions, approval rules and run log underneath are the same harness in every case.
- Guidely / Neutral guide, docked assistant
- Jarvis / Voice-led, in the search bar
Dialogue
Docked dialogue panel
A conversation with the context header on top and the run log in its own sidebar. Best for guided work, onboarding and occasional users who need the product explained while it is operated for them.
- New and occasional users
- Multi-step guided workflows
- Talk or type, same thread
Wand
Keyboard-first wand
A command palette on the pattern power users already know from developer docs search. Ask for an outcome, get navigation, actions and answers in one list, run it without leaving the keyboard.
- Power users who want speed
- Jump to any screen or record
- Dictate the request if faster
Character
An animated character on the page
A small character that sits in the corner of your product, notices where the user is, points at the row or field it is talking about and opens a bubble with the next useful move. Clicking it expands into the full dialogue.
- Training-heavy workflows
- Users with low tech confidence
- Nudges instead of a support call
In both form factors the run log lives on Guidely's own surface: a rail beside the conversation or attached to the palette. Actions are written back through your APIs and audit trail, not pinned into your product chrome.
What users can point it at
Six things Guidely does inside a mature product
Each one is a product engineering problem before it is a model problem, which is why Guidely is built against your workflows rather than dropped in as a generic assistant.
Read the context
Account, module, record, role and open obligations, resolved from live product data rather than a help centre article.
Suggest the next move
Two or three things worth doing here, ranked by what is overdue, at risk or contractually due.
Walk it through
A guided run through the workflow with plain-language explanation in the customer's own domain terms.
Do the mechanical steps
Multi-step actions across records, queues and integrations, executed through your product's own APIs and permissions.
Draft the work product
Notices, summaries, reports and follow-up messages in your customers' formats, ready for review.
Hand off cleanly
When confidence is low or policy says stop, Guidely routes to the right human with context already assembled.
Worked example
What one request actually triggers
"Chase every overdue inspection in the Northeast and get the escalations out today" is five steps of product work, not a chat reply.
- Step 1
State the context
Northeast region, 14 overdue inspections, signed in as a compliance manager.
- Step 2
Show the options
Remind the inspectors, escalate to the client, or reschedule. Guidely explains the trade-off of each.
- Step 3
Walk the workflow
Groups the sites by inspector, drafts 9 reminders and 5 client escalation notices for review.
- Step 4
Ask for the one decision
Two sites breach a contractual SLA, so Guidely asks a manager before anything reaches the client.
- Step 5
Do it and log it
Sends, updates statuses, files the notices and writes every action to the audit trail.
Hand-holding your customers
Guidely teaches your product while it operates it
Most vertical SaaS products carry more capability than the average user ever reaches. Guidely closes that gap without another training programme.
Guided completion
- Fills every field it can infer, asks only for real decisions
- Explains what each step means in the customer's own domain language
- Shows the change before it is committed, so trust builds over time
- Points to the module that solves the problem instead of hiding it in a menu
What that changes commercially
- Adoption
- Premium modules get used because Guidely introduces them at the moment they solve the user's problem.
- Support load
- Fewer how-do-I tickets and onboarding calls, because discovery and guidance live inside the workflow.
- Expansion
- A defensible reason to move a customer to a higher tier, and a differentiator against AI-native entrants.
- Retention
- Your product stops looking dated in renewal conversations where AI is now part of the checklist.
Where it plugs in
Built into the product you already have
Guidely is the visible front of an agent harness. That is what lets it take real actions in a permission-heavy product without becoming a liability.
Its own surface, your product
Guidely renders as a named panel docked in your application, in your design system. No new portal and no second login, but clearly its own layer.
Uses your data model
It reads and writes through your services and permissions, so what a user can do with Guidely is exactly what they could already do by hand.
Runs on the harness
Intent, context, tools, approval, action and audit. Guidely is the visible front of a governed system, which is why it can be trusted with real actions.
No rebuild
The first useful Guidely ships against the product you have today, in weeks, alongside your roadmap rather than instead of it.
This is one reference pattern for supervised AI inside an established SaaS product. We deliver it inside a capability wave, against the product you already own.
Your customers already expect AI from your product.
Let us make it real without rebuilding what already works.