Inventory management on Claude
Claude, Codex, Cursor, OpenCode, a framework your team wrote, something you host yourself because the data cannot leave the building: all of them are welcome, and none of them need a plugin. We do not sell you an agent, because the one you already trust should be the one running your inventory.
What makes a system agent-ready
Not a chat box bolted onto a sidebar. An agent needs to find out what it can do, be trusted with a limited amount of it, know when something happens, and not break when we ship on Thursday.
A manual written for agents
One markdown file per area of the product, rewritten on every release so it cannot fall behind what the product actually does. An agent reads the two files it needs rather than all of them, and a written brief covers the things a reference cannot: how access works, where the rate limit bites, and the mistakes that cost somebody an afternoon.
Give it a key that cannot do everything
Permissions and scopes are set per key. An agent that drafts purchase orders gets a key that drafts purchase orders, and cannot touch your pricing or delete a customer. You generate it, you revoke it, and you do not have to trust a model with the keys to the building.
It can ask what changed
Lists are filterable by update time, so an agent can ask what moved since it last looked instead of re-reading your catalogue. Webhooks are on the roadmap and not in v1 yet, so polling on a sensible interval is the current answer, and for most agent work it is the cheaper one anyway.
Everything, not a read-only export
Coverage is close to total: the same sale orders, stock, purchase orders, shipments, invoices, customers and suppliers a person works with. An agent that can only read writes you a report. One that can act finishes the job.
Nothing bespoke to teach it
REST, resource-oriented URLs, JSON in and JSON out, standard status codes, the same filtering and pagination on every list. Models are good at this because everything else is like it. A proprietary protocol is a thing you would have to explain in every prompt.
We will not break it on a Thursday
The API evolves additively under `/v1`: new endpoints and fields appear, nothing is quietly removed or renamed. Build the agent to ignore what it does not recognise and it keeps working. A breaking change gets a new version in the path, so you move when you are ready.
One prompt to get it started
Nothing to install and no SDK to learn. The toolkit is markdown and a handful of scripts: point your agent at it, let it read the brief, and it knows the whole surface before it makes a single call.
> Clone the Qoblex agent toolkit from https://github.com/Qoblex/agent-toolkit, read AGENTS.md, and tell me what you can do here.· Read it. Auth is one header, and there is a rate limit worth respecting. The reference covers 201 endpoints: stock, purchase orders, sale orders, shipments, invoices, products and contacts. Set QOBLEX_API_KEY and I am ready to work.The key is yours to issue and revoke: in Qoblex, Integrations, then the API tile. Scope it to what the agent needs and nothing more. The reference ships inside the toolkit as markdown, one file per resource group, which is what an agent reads rather than the raw OpenAPI document.
What it looks like with an agent driving
Two sessions, run against a real account. Nobody wrote an integration for either: the agent read the spec, chose the endpoints and made the calls. What you type is the question you would have asked a colleague.
> Which products are running low and sold well over the last two months? Draft the purchase orders to restock them.GET /v1/reporting/forecasting
?from_date=2026-07-22&to_date=2026-09-20
&coverage_target_weeks=4· 27 of 67 variants are projected to run short. They come from 5 suppliers, so that is 5 draft purchase orders for you to approve.> Which batches expire in the next 90 days, and how much stock is in them?GET /v1/batches
?filters=expires_at<=2026-12-19
&sort_by=expires_atGET /v1/batches?…&page=1GET /v1/batches?…&page=2· 88 batches expire inside 90 days, holding 11,267.19 units. 78 of them are already past expiry today.Both sessions are reads. The API can ship stock and raise invoices too, and an agent can drive that, but those move goods and money: the toolkit tells an agent to get a human to confirm first, and so do we.
Frequently asked questions
Agents, access, safety, and what we do and do not provide.
Can an AI agent manage inventory in Qoblex?
Yes. We publish a manual written for agents, so one can find out what it is allowed to do and then do it: check stock, draft and raise purchase orders, allocate and ship sale orders, raise invoices, update products. You bring the agent and Qoblex provides the surface it drives.
Which agents work with it?
Any agent that can read a markdown file and make a web request, which in practice is all of them. Claude, Codex, Cursor, OpenCode, OpenClaw, a framework your team built, or something self-hosted on your own hardware. We do not certify a list, because a list would go stale faster than this page.
Do you sell an AI agent or a copilot?
No, and not by accident. An agent rented from your inventory vendor is limited to that vendor’s data and that vendor’s pace. We would rather make Qoblex something any agent can drive properly, and leave the choice of agent to you.
How do I stop an agent doing something it should not?
Give it its own API key and scope that key to what it needs. Permissions are set per key, so an agent that reorders stock cannot edit pricing, and one that reads reports cannot write anything at all. Revoke the key and it stops, immediately, without touching anyone’s login.
Does an agent have to poll for changes?
For now, yes. Webhooks are on the roadmap and are not in the v1 API yet, so an agent checks in rather than being called. That is less of a limitation than it sounds: list endpoints filter on update time, so a poll returns what changed rather than everything, and an agent on a few-minute interval keeps up with most inventory work. See the API reference for the current state.
Can I run the agent on our own infrastructure?
Yes. Nothing about this requires a hosted anything: the agent runs wherever you put it and talks to the API over HTTPS. For teams whose data cannot leave their own network, the agent stays inside it and only the API calls go out.
Will an update break our agent?
It should not. The API evolves additively under `/v1`, so new endpoints and fields can appear but nothing is quietly renamed. Build the agent to ignore fields it does not recognise. Anything genuinely breaking gets a new version in the path.
What does it cost?
There is nothing to buy for the agent, because we are not selling one. What it needs is API access, which is a paid add-on on Starter and Business and included on Scale. See pricing.
The rest of the developer story
How it works
Be up and running in a fraction of the time you’re expecting it to take.
- 1
Talk to an expert
Our experts have helped hundreds of product sellers just like you.
Book a demo - 2
Bring in a specialist
Engage an onboarding specialist to make it much easier for you.
Book a demo - 3
Go live fast
Unlike other systems, we can get you up and running within 1 week.