Custom AI tools, working in weeks.
Prasinus builds AI tools for businesses without an AI team. They take over the calls, emails, and paperwork your people handle by hand, run in your own accounts, and stay yours.
Six tools most businesses can use.
Every company has a pile of work people handle by hand: calls nobody has time to listen to, emails and PDFs to re-type, customers to sort, answers to dig up.
These are the tools we build most. Each one is built around your process and your data, and anything else follows the same pattern.
-
Call Insights
Every recorded call is transcribed, checked against your standards, and summarized. Unhappy customers, missed bookings, and promised callbacks land in a manager’s inbox the next morning.
Works fromRecordings from your phone system
-
Customer & order intelligence
Every customer, order, product, or photo is tagged automatically by industry, category, or whatever you define, so you can see who buys what and where to sell next.
Works fromCustomer lists, order history, catalogs
-
Inbox & paperwork autopilot
Emails, PDFs, RFQs, purchase orders, and forms are read, checked against your records, and drafted into your system. A person approves every entry before it goes in.
Works fromShared inboxes, attachments, forms
-
Company brain
Your team asks a question and gets the answer from your own SOPs, manuals, specs, and past jobs, with the source attached. If the answer isn’t there, it says so instead of guessing.
Works fromSOPs, manuals, specs, past quotes
-
Equipment photo reader
The nameplate photos your techs already take are read and filed under each customer: brand, model, serial, and age. Aging, phased-out, and recalled equipment becomes a list of customers worth a call.
Works fromJob photos in your field-service app
-
Signal watch
Every night it checks the companies you sell to for public news: expansions, new leaders, hiring, awards. Each item is checked against its source, and your reps get a morning email with a reason to call.
Works fromYour account list and public sources
See all six working.
Short screen recordings of each tool, captioned so they work with the sound off.
The companies, people, and documents in the first five are made up. Signal watch checks real public companies, shown as sample accounts of a made-up distributor. Everything on screen is the tool’s real output, unedited.
What a manager sees at 7 AM.
Every call from yesterday was transcribed and reviewed overnight. These are the few that needed a person.
Real output from our Call Insights demo, unedited. The company and its calls are fictional, written and voiced for the demo.
| Call | Flagged for | Why it matters |
|---|---|---|
| 4:50 PMFurnace not kicking on |
| Eric had no heat and wanted service, but declined after hearing the after-hours surcharge and no appointment was booked. |
| 9:40 AMCancel no-heat appointment |
| Customer cancelled a no-heat appointment and switched to AirPro due to price. |
| 10:25 AMFurnace not making heat |
| Harold said he was upset about not knowing when to be ready and ended resigned to sitting by the phone. |
| 8:15 AMFurnace not starting |
| Caller had no heat, wanted same-day service, and agreed to hear about Comfort Club. |
Ask for the live demo Twenty minutes, at your office or on video. See the briefing, search the calls, and hear the moments that got flagged.
The AI infrastructure underneath, set up in your accounts.
Every tool runs on infrastructure we set up and hand over. If your team already uses AI and it needs a proper home, we build that on its own.
This is the part we know best: the plumbing that keeps AI working long after the demo, built in your cloud and documented so your team can run it.
Proof you can check
The six demos share one foundation.
One model layer that switches between AI providers, speech-to-text and document search running on the machine itself, a cost ledger for every model call, and a scorecard for each tool against an answer key.
- Model access
- Your accounts, your terms. Azure OpenAI, AWS Bedrock, Google Vertex AI, or the Anthropic and OpenAI APIs directly, behind one layer, so switching providers is a setting, not a rewrite.
- Private models
- For data that can’t leave. Speech-to-text, search, and smaller language models running on your own servers or GPUs.
- Document search
- Answers from your own files. SharePoint, Drive, or file shares indexed for search, with access that follows who’s asking.
- Cost tracking
- Every model call on the ledger. Cost per tool, per team, and per item, with budgets and alerts before a bill surprises anyone.
- Testing
- Scored before it ships. A test set your team labels, rerun on every change and every model upgrade, so quality is a number, not a feeling.
- Running it
- Watched and recoverable. Health checks, error alerts, a log of who asked what, keys in a vault, single sign-on, and a manual fallback for when the AI is down.
Set up on its own, it’s a First Build: fixed price from $7,500, three to four weeks, and yours to keep.
Three steps, each easy to say yes to.
Start with a free look at your business, see one tool working on your own data, then keep building if it’s worth it.
Fixed prices · no long contracts
Working in weeks, not quarters.
You see a sample of your own data running through it in week one. You pay the second half when it passes a checklist we write together.
- Opportunity Scan
- Free · 60 minutes. We walk through your departments with the people doing the work, then send a one-page list of the three things most worth building, with rough value.
- First Build
- From $7,500 fixed · 3–4 weeks. One working tool on your real data, in your accounts. Half to start, half when it passes the checklist.
- AI Partner
- From $1,500 a month. We keep your tools running and build the next one. Plans at $1,500, $6,000, and $9,500; three months, then month to month.
- Running costs
- AI and hosting are billed to your own accounts, usually $20–$300 a month per tool. No markup and no lock-in.
What you keep
- WorkingA tool doing real work within weeks, measured against a before number.
- YoursThe code in your GitHub, the configuration, a runbook, and a recorded walkthrough.
- PrivateYour data stays in your accounts. We keep no copies.
- NextA short list of what to build next, and help building it if you want it.
From first conversation to a tool your team uses.
The same five steps every time, so you always know what happens next and what you get.
-
01
Look at the work
We sit with the people who do the work, count the volume, and find the piles: calls, emails, documents, and lists handled by hand.
ResultThe Opportunity Scan one-pager
-
02
Pick one tool
Choose the opportunity with clear value and data you already have, and write down what “working” means before anything is built.
ResultA fixed price and a pass/fail checklist
-
03
Build it on your data
A sample of your own calls or documents runs through it in the first week. The rest of the build is tuning it with the people who will use it.
ResultA working tool, checked against real examples
-
04
Put it to work
It runs in your accounts with monitoring and a manual fallback. A person approves anything that goes to a customer or into your systems.
ResultA tool in daily use
-
05
Keep it, and grow it
You get the source, the runbook, and training. Keep us on to run it and build the next one, or run it yourselves.
You keepThe capability, not a dependency
See the full working process What Prasinus and your team each do, and what you have at the end of each step. Published under CC BY 4.0.
A shared build, with clear ownership.
Prasinus brings the AI and software work. Your team brings the judgment that makes the tool useful.
| Step | Prasinus brings | Your team brings | Result |
|---|---|---|---|
| Look | Questions, outside perspective, rough value math | An hour and the people who do the work | Three opportunities, ranked |
| Pick | Scope, approach, and a fixed price | A decision and an owner | A checklist for “working” |
| Build | Software, AI, data, and testing | Sample data, access, and feedback | A working tool |
| Run | Setup in your accounts, monitoring, fixes | Users and approvals | A tool in daily use |
| Keep | Source, runbook, and training | A named owner | A capability you own |
Common questions
Where does our data go?
It stays in your accounts: your cloud, your AI provider account, your storage. We work through logins you control and can revoke, and we keep no copies.
What if we stop working together?
Everything keeps running. The code, configuration, and runbook are yours, and any competent developer can pick them up. That is on purpose.
Does the AI make decisions on its own?
Not unless you decide it should, after you have watched it work. By default a person approves anything that goes to a customer, commits a price, or changes a record.
We already have ChatGPT or Copilot. How is this different?
Those help one person at a time through a chat box. We build tools that work through the whole pile every day, every call or every document, and put the results into your process.
Is there a good place to start?
Answer for one part of the business. “Not sure” is a useful answer; it just tells us what the first conversation needs to cover.
Let’s talk
Tell us what your team spends too much time on. We reply within one business day.
me@logan.placeNo deck required. Please do not send confidential data, files, or credentials in the first email.