Read-only API
Endpoints that can only read.
A versioned JSON API on top of the distributor’s existing business application, with product and customer endpoints that can only read.
Use case · AI and automationAsk the ERPa question andget an answerfor your branch
CBOS built a read-only link between a distributor’s ERP and warehouse system and Y Intelligence, our AI platform. Staff ask about products, stock and customers in plain language; the assistant can only read, and each person sees only the branches they’re allowed to see.
6read-only tools the assistant can use3regional branches, each scoped to its own staff
The problem
Every customer question turns into a hunt through the ERP
A customer phones to ask whether you’ve got 40 units and what they still owe. To answer, someone logs into the ERP, finds the stock screen, checks they’re in the right branch’s database, then opens the account on another screen. Staff who’ve learnt every screen answer quickly; everyone else asks them.
Customers wait on hold
The rep puts the caller on hold to click through stock, pricing and account screens. Long calls mean fewer calls answered, and some buyers ring the next supplier instead.
The wrong branch’s numbers
Each branch keeps its own database, so staff have to pick the right one before they search. Pick the wrong one and you quote stock you don’t have.
One person who knows the system
Questions queue up for whoever knows the ERP best. They lose a good part of the day to interruptions, and when they’re on leave the answers slow right down.
Account details in pieces
Balance on one screen, open orders on another, contact details on a third. Miss one and a rep promises delivery to an account that’s over its limit.
Weeks before new staff can answer
New starters learn menus and report names before they can answer a basic question, so for their first weeks they lean on the people around them.
AI that IT won’t allow near the ERP
A general AI tool with a database login could read every branch and change records, so the request gets turned down. Meanwhile staff paste exports into public chat tools anyway.
Your systems already hold the answers. Your team needs a quick way to ask for them, and one that IT is happy to switch on.
Before and after
From searching screens to asking a question
Flip the switch to see the same questions answered with the assistant in place.
BeforeWith the system
- Staff log in and search ERP screens for each answerStaff ask the assistant a question in plain language
- Knowing which branch’s database to checkThe assistant queries the right branch for the person asking
- Customer details pieced together from several screensProduct and customer answers come back from one question
- Questions queue up for the one person who knows the systemAnyone with access asks the assistant directly
- An AI tool would need a login that sees everythingEach request runs as the person asking and sees only what they may see
- AI tools kept away from live business dataThe assistant reads live data through a locked-down, read-only API
Four screens, the right branch database and a colleague to ask, for one stock figure.
One question in plain language, answered from your branch’s live data.
How it works
Four layers between the question and the data
The assistant only reaches the ERP through these four pieces of work. Click through each one.
01A read-only API on the ERP
02A key on every call
03Runs as the person asking
04Six tools in Y Intelligence
Step 01
A read-only API on the ERP
We put a versioned JSON API on top of the distributor’s existing business application. Its product and customer endpoints can only read, so there’s no route through it that changes a record.
Step 02
A key on every call
Every call to the API needs a key. A command-line tool issues and manages the keys, so nothing reaches the data without one.
Step 03
Runs as the person asking
Each request goes through a per-user token exchange and runs as the person who asked. A branch user’s answer covers their branch; someone allowed to see the whole group gets the whole group.
Step 04
Six tools in Y Intelligence
The connector pack gives the assistant six read-only tools for products, stock and customers, by region or for the whole group. It picks the tool that fits the question and answers from the data as it stands at that moment.
What we built
Read-only access, scoped to each person
API keys and a key tool
Nothing gets in without a key.
Every call needs an API key, and a command-line tool issues and manages the keys.
Branch scoping
Answers cover the branches you may see.
Each request runs as the person asking, through a per-user token exchange, so answers only cover the branches that person is allowed to see.
Y Intelligence connector pack
Six tools the assistant can call.
Six read-only tools that let the assistant look up products, stock and customers across the regions, or for the whole group.
Live answers
The data as it stands when you ask.
The assistant answers from the ERP and warehouse data as it stands at the moment someone asks.
Hover over a card for more.
What changes
What changes for the people doing the job
Your ERP stays as it is. More of your team can get an answer out of it without asking the person who knows the screens.
Sales and account staff
Answer a stock or account question while the customer is still on the line, in the words the customer used.
Whoever knows the ERP best
The “where do I find” questions go to the assistant, and their own work gets done.
New starters
Ask in plain language from the first week, before they know where every screen lives.
IT and the owners
AI on live business data with read-only access, a key on every call and answers scoped to the person asking.
Questions we get
Before you let AI near your ERP
What IT and the owners will want to know before saying yes. Open each one.
Can the assistant change anything in our ERP?
No. The API only has read endpoints for products and customers, and every call needs a key. There’s no route through it that writes to the ERP.
Do we have to replace our ERP?
No. The API sits on top of the business application you already run, and the ERP and warehouse system carry on as they are.
Will staff see other branches’ numbers?
Only if they’re allowed to. Each request runs as the person asking through a per-user token exchange, so the answer covers the branches that person may see, or the whole group for someone with that access.
How current are the answers?
The assistant reads the ERP and warehouse data at the moment someone asks, so the answer reflects the data as it stands then.
Where would we start?
With a conversation about the questions your team asks most and who should see what. A first release answers those questions read-only, and more tools follow once they’ve proved useful.
Built with
The stack
PythonFlaskSQL ServerREST and JSONAPI-key authenticationY Intelligence connector packs
Quick check
Sound like your business?
Tick what sounds familiar.
Your staff ask colleagues for stock or customer details because the ERP is hard to search.
You run more than one branch or region, each with its own data.
You’d like to use AI on your business data but don’t want it to be able to change anything.
Different people should only see the data for their own branch or accounts.
New staff take weeks to learn where everything lives in your systems.
of 5 ticked
0 to 1 ticked
Your team probably finds what it needs without much trouble. If that changes as you grow, we’d be glad to help.
2 to 3 ticked
Some of your team’s time goes into hunting for answers your systems already hold. A read-only assistant on top of them could be a small, safe first step.
4 to 5 ticked
Talk it through with us
Your data is there, but getting answers out of it is slow and access is hard to control. We’ve built this before, so let’s talk about your setup.
More case studies
All case studies Other businesses we’ve built for
Let's talk
Want AI on your business data without handing over the keys?
Tell us which systems your team searches every day and who should see what, and we’ll show you how a read-only assistant would work for you.
Use case · AI and automationAsk the ERPa question andget an answerfor your branch
CBOS built a read-only link between a distributor’s ERP and warehouse system and Y Intelligence, our AI platform. Staff ask about products, stock and customers in plain language; the assistant can only read, and each person sees only the branches they’re allowed to see.
6read-only tools the assistant can use3regional branches, each scoped to its own staff
The problem
Every customer question turns into a hunt through the ERP
A customer phones to ask whether you’ve got 40 units and what they still owe. To answer, someone logs into the ERP, finds the stock screen, checks they’re in the right branch’s database, then opens the account on another screen. Staff who’ve learnt every screen answer quickly; everyone else asks them.
Customers wait on hold
The rep puts the caller on hold to click through stock, pricing and account screens. Long calls mean fewer calls answered, and some buyers ring the next supplier instead.
The wrong branch’s numbers
Each branch keeps its own database, so staff have to pick the right one before they search. Pick the wrong one and you quote stock you don’t have.
One person who knows the system
Questions queue up for whoever knows the ERP best. They lose a good part of the day to interruptions, and when they’re on leave the answers slow right down.
Account details in pieces
Balance on one screen, open orders on another, contact details on a third. Miss one and a rep promises delivery to an account that’s over its limit.
Weeks before new staff can answer
New starters learn menus and report names before they can answer a basic question, so for their first weeks they lean on the people around them.
AI that IT won’t allow near the ERP
A general AI tool with a database login could read every branch and change records, so the request gets turned down. Meanwhile staff paste exports into public chat tools anyway.
Your systems already hold the answers. Your team needs a quick way to ask for them, and one that IT is happy to switch on.
Before and after
From searching screens to asking a question
Flip the switch to see the same questions answered with the assistant in place.
BeforeWith the system
- Staff log in and search ERP screens for each answerStaff ask the assistant a question in plain language
- Knowing which branch’s database to checkThe assistant queries the right branch for the person asking
- Customer details pieced together from several screensProduct and customer answers come back from one question
- Questions queue up for the one person who knows the systemAnyone with access asks the assistant directly
- An AI tool would need a login that sees everythingEach request runs as the person asking and sees only what they may see
- AI tools kept away from live business dataThe assistant reads live data through a locked-down, read-only API
Four screens, the right branch database and a colleague to ask, for one stock figure.
One question in plain language, answered from your branch’s live data.
How it works
Four layers between the question and the data
The assistant only reaches the ERP through these four pieces of work. Click through each one.
01A read-only API on the ERP
02A key on every call
03Runs as the person asking
04Six tools in Y Intelligence
Step 01
A read-only API on the ERP
We put a versioned JSON API on top of the distributor’s existing business application. Its product and customer endpoints can only read, so there’s no route through it that changes a record.
Step 02
A key on every call
Every call to the API needs a key. A command-line tool issues and manages the keys, so nothing reaches the data without one.
Step 03
Runs as the person asking
Each request goes through a per-user token exchange and runs as the person who asked. A branch user’s answer covers their branch; someone allowed to see the whole group gets the whole group.
Step 04
Six tools in Y Intelligence
The connector pack gives the assistant six read-only tools for products, stock and customers, by region or for the whole group. It picks the tool that fits the question and answers from the data as it stands at that moment.
What we built
Read-only access, scoped to each person
Read-only API
Endpoints that can only read.
A versioned JSON API on top of the distributor’s existing business application, with product and customer endpoints that can only read.
API keys and a key tool
Nothing gets in without a key.
Every call needs an API key, and a command-line tool issues and manages the keys.
Branch scoping
Answers cover the branches you may see.
Each request runs as the person asking, through a per-user token exchange, so answers only cover the branches that person is allowed to see.
Y Intelligence connector pack
Six tools the assistant can call.
Six read-only tools that let the assistant look up products, stock and customers across the regions, or for the whole group.
Live answers
The data as it stands when you ask.
The assistant answers from the ERP and warehouse data as it stands at the moment someone asks.
Hover over a card for more.
What changes
What changes for the people doing the job
Your ERP stays as it is. More of your team can get an answer out of it without asking the person who knows the screens.
Sales and account staff
Answer a stock or account question while the customer is still on the line, in the words the customer used.
Whoever knows the ERP best
The “where do I find” questions go to the assistant, and their own work gets done.
New starters
Ask in plain language from the first week, before they know where every screen lives.
IT and the owners
AI on live business data with read-only access, a key on every call and answers scoped to the person asking.
Questions we get
Before you let AI near your ERP
What IT and the owners will want to know before saying yes. Open each one.
Can the assistant change anything in our ERP?
No. The API only has read endpoints for products and customers, and every call needs a key. There’s no route through it that writes to the ERP.
Do we have to replace our ERP?
No. The API sits on top of the business application you already run, and the ERP and warehouse system carry on as they are.
Will staff see other branches’ numbers?
Only if they’re allowed to. Each request runs as the person asking through a per-user token exchange, so the answer covers the branches that person may see, or the whole group for someone with that access.
How current are the answers?
The assistant reads the ERP and warehouse data at the moment someone asks, so the answer reflects the data as it stands then.
Where would we start?
With a conversation about the questions your team asks most and who should see what. A first release answers those questions read-only, and more tools follow once they’ve proved useful.
Built with
The stack
PythonFlaskSQL ServerREST and JSONAPI-key authenticationY Intelligence connector packs
Quick check
Sound like your business?
Tick what sounds familiar.
Your staff ask colleagues for stock or customer details because the ERP is hard to search.
You run more than one branch or region, each with its own data.
You’d like to use AI on your business data but don’t want it to be able to change anything.
Different people should only see the data for their own branch or accounts.
New staff take weeks to learn where everything lives in your systems.
of 5 ticked
0 to 1 ticked
Your team probably finds what it needs without much trouble. If that changes as you grow, we’d be glad to help.
2 to 3 ticked
Some of your team’s time goes into hunting for answers your systems already hold. A read-only assistant on top of them could be a small, safe first step.
4 to 5 ticked
Talk it through with us
Your data is there, but getting answers out of it is slow and access is hard to control. We’ve built this before, so let’s talk about your setup.
More case studies
All case studies Other businesses we’ve built for
Let's talk
Want AI on your business data without handing over the keys?
Tell us which systems your team searches every day and who should see what, and we’ll show you how a read-only assistant would work for you.
Let's Connect
Interested in building a smarter, more connected operation? Contact us to discuss how AI can support your business goals.