AI features added to the product you already run.
A copilot, search with sources, document reading or voice notes, built into your existing app and tested on real questions before your users see it. Your team keeps shipping while we add it.
Not the right team for a chatbot added to a page with no access to your data. For a new app from scratch, see AI web apps.
- Pilot
- 4 to 6 weeks
- From
- US$2,400
- Built into
- Your code and stack
Hi Dana, sorry about that. I can see two charges on 1 November, so I've queued a refund for the second one. It will reach your card in 3 to 5 working days.
Eight AI features we add to existing apps.
Each one links to a use case with how it works, a price range and a timeline. Most products start with one.
A copilot in your product
An assistant inside your software that answers from your own help docs and data.
Use case: support assistant →Search with sources
Ask across your documents and get an answer that shows where it came from.
Use case: company search →Document reading
PDFs and scans read into your system, with anything uncertain sent to a person first.
Use case: document extraction →Voice notes
Speak a note and get a drafted record, invoice or report to check.
Use case: voice-note invoices →Support triage
Every ticket sorted and answered in draft from your help articles. A person sends each reply.
Use case: inbox triage →Ask your data
Plain-English questions over your database, answered with the numbers and the query behind them.
Use case: ask your data →A quality gate
Every AI output checked against your rules before it reaches a customer.
Use case: quality gate →Your app in Claude or ChatGPT
Your product's data and actions, available inside the AI chat your customers already use.
Use case: connect to Claude →AI features in daily use.
An AI scribe inside Tibra Health's clinical app, and support AI added to the platform we rebuilt for Parallel Completions.
- 01Blueprint
- 02Build
- 03Live
Hi Dana, sorry about that. I can see two charges on 1 November, so I've queued a refund for the second one. It will reach your card in 3 to 5 working days, and your plan stays the same.
- 01Blueprint
- 02Build
- 03Live
Illustrations of the two builds · sample data
An AI feature, one priced step at a time.
Each step has its own price and its own result, so you can stop after any of them and keep what you paid for.
Scoping call
30 minutes on your product, your users and where they lose time.
- Time
- 30 minutes
- Price
- Free
- Ends with
- A written next step
AI blueprint
Your AI ideas ranked by return and effort, a clickable prototype of the feature, and a fixed estimate.
- Time
- 1 to 2 weeks
- Price
- US$1,800 fixed
- Ends with
- A plan you own, whoever builds it
Pilot
One AI feature in your app on your own data, measured against the task it replaces.
- Time
- 4 to 6 weeks
- Price
- From US$2,400
- Ends with
- A go or stop on the full release
Version 1
The feature released to all your users, with tests on real questions and a backup model.
- Time
- 5 to 6 weeks a phase
- Price
- From US$7,200 a phase
- Ends with
- The feature live for every user
Run and grow
Accuracy checks, model updates and new features as your users ask for them.
- Time
- Monthly
- Price
- From US$599 a month
- Ends with
- A report every month
In US dollars · monthly plans on the pricing page
Built into your stack, not beside it.
We work in your repository and your cloud, and leave each feature documented for your team.
Your repo and your stack
The feature is built in your codebase and goes through your own reviews, so your team can run it.
We need access, and a reviewer on your side.
Answers from your own sources
Search and copilots answer from your documents and data and show the source, or say they don't know.
Your documents have to be kept up to date.
Tests on real questions
A set of real questions with known answers runs before every release.
Someone on your side checks the answers at the start.
A main model and a backup
Each feature uses the model that does its job best, and the app says when the backup answered.
Two providers to watch for price and policy changes.
Model costs you can see
You pay what the model providers charge, with no markup, under a monthly cap you set.
Heavy use means a bigger model bill, so the cap needs watching.
A written context file
Every repo explains itself, so any engineer or AI assistant can pick up the code where we left it.
It has to be updated with every change.
How each feature stays right.
The same guardrails go into every AI feature we ship, whatever it does.
People approve
Replies, notes and records stay drafts until a person approves them.
How we build →A confidence bar
Anything the AI isn't sure about goes to a person, and the bar moves if it drifts from real outcomes.
Quality gate →A backup model
If the main model fails, a backup answers, and the app tells the user.
How we build →Read-only where it should be
Tools that look at your code or data can read it and never write to it.
How we build →Tested before release
The same set of real questions runs before every change ships.
Quality gate →The AI only sees what each user can.
A feature inside your product follows your product's access rules. Security starts in the first phase, not after launch.
- Access
- AI search only reads what the person asking is allowed to open.
- Where it lives
- In the region your rules require. The clinical platform we run keeps its data in the EU.
- Model providers
- Your data goes only to the providers named in the AI blueprint, for the feature that needs it.
- Your code
- Tools that read your codebase, such as bug escalation, can read it and never write to it.
- Reviews done
- An ISO 27001 and GDPR review for Parallel Completions, and a security brief for Tibra Health's clinical AI app.
What people ask first.
Short answers here. The scoping call gets you the long ones for your product.
Do we have to rebuild our app?
No. The feature goes into your existing codebase and your current release process.
What does an AI feature cost?
A pilot starts from US$2,400 and a full release from US$7,200 a phase. Model costs are passed on with no markup, under a monthly cap you set.
Which models do you use?
The one that does each job best, named in the AI blueprint, with a backup from a second provider.
Can our developers work with you?
Yes. We work in your repo through changes your team reviews, and leave a written context file so they can run the feature.
How do we know it's working?
The pilot is measured against the task it replaces, and every month you get a report on how it performed and what it cost to run.
What if the AI gets something wrong?
People approve what it writes before anything is saved, every feature is tested on real questions before it ships, and a backup model takes over if the main one fails.
Get this scopedin 30 minutes.
Bring your product and the task you want the AI to take on. You leave with a written next step and a price range.
India time · confirmed on Cal ID