Next.js 16 and React 19
The workspace and the public site share one application, server-rendered, with the marketing pages statically generated.
AI-NATIVE OPERATIONS SOFTWARE
Inventory software that
answers for itself.
Most inventory systems can tell you what the number is. Dino Inventory can tell you which records produced it, which specialist read them, and which policy applies — in plain language, on live data, inside the permissions of the person asking.
See the application- 98
- database tablesThe operational schema behind purchasing, inventory, fulfillment, finance, and manufacturing.
- 93
- application screensEach module has its own register, detail view, and workflow actions.
- 6
- AI specialist agentsA supervisor routes each question to the agent that owns that domain.
- 896
- cross-browser testsPassing on Chromium, Firefox, and WebKit in a single suite run.
THE DIFFERENCE
AI that reads the ledger, not a copy of it.
The assistant holds no separate index of your operations and no elevated credentials. A supervisor routes each question to one of six domain specialists — inventory, procurement, sales, finance, manufacturing, policy — and each calls the same API endpoints the interface calls, authenticated as the person asking. An answer is therefore as current and as permitted as the screen beside it.
How the assistant works- Six specialists with disjoint domain ownership
- Live reads under the asker's own permissions
- Tool calls and records visible with every answer
- Policies cited by document title, or declared absent
- Any model provider, with per-capability fallbacks
HOW IT IS VERIFIED
Evidence, not assurances.
Inventory software earns trust by being correct about quantities. Every claim on this page is backed by a suite that runs on demand — including invariants asserting that balances reconcile to the ledger and that a transfer never creates or destroys stock.
- 896 end-to-end tests across 22 suites, run on three browser engines.
- 33 database invariants checked directly against the live schema.
- Load, stress, spike, and endurance profiles — roughly 16,000 requests with no errors.
- Accessibility auditing against WCAG 2.1 A and AA with axe-core.
- Backup, restore, and verification scripted across all three data stores.
ARCHITECTURE
Built to be replaced in parts,
not rewritten whole.
No layer assumes a specific vendor. The AI provider, the model, and the vector store are configuration, not architecture.
Node and MySQL
A single API service over a relational schema, where the inventory ledger is the source of truth and balances reconcile to it.
LangGraph multi-agent supervisor
Six domain specialists with disjoint module ownership, each calling the same API endpoints the interface uses.
Qdrant vector search
Policies and procedures are embedded and searched by meaning, so the assistant can cite the document it used.
Provider-agnostic AI
Models are configured per capability with backups. No model vendor is embedded in the product.
WHERE IT STANDS
Honest about the line between built and deployed.
The application is feature-complete across its modules and runs today as a containerized stack with demonstration records. What remains is deployment: hosting, certificates, mail delivery, and provider credentials. Those are operator decisions, and this page does not claim them as finished.
Questions and answers- Built: every module, the AI assistant, the verification suite
- Built: backup, restore, and integrity tooling
- Operator's call: hosting, TLS, and domain
- Operator's call: AI provider credentials and spend
FOR A CLOSER LOOK
Questions worth asking.
What makes Dino Inventory AI-native rather than AI-added?
The assistant is not a chat window bolted onto a database. A supervisor routes each question to one of six domain specialists, each of which reads live records through the same API and the same permissions as the person asking. The AI has no separate data copy and no elevated access, so it cannot drift from the operational truth or leak across permission boundaries.
How is the AI prevented from inventing figures?
Each specialist is scoped to the modules it owns and answers from records it has read, with the tool calls and the records returned visible alongside the answer. The policy agent quotes the document title it used and states plainly when no document covers a question rather than inferring a rule.
Which AI provider does the platform depend on?
None structurally. Providers are configured by the operator, and models are chosen per capability — chat, embeddings, reranking, speech, and transcription — each with a backup. A workspace can run entirely on free models and move to paid models without other changes.
What stage is the product at?
The application is feature-complete across its modules and verified by an automated suite that runs on three browser engines. It runs today as a containerized stack with demonstration data. Production deployment decisions — hosting, TLS, mail delivery, and provider credentials — belong to the operator and are not published here.
Are the figures on this page audited?
They describe the software, not a business. Counts of tables, screens, agents, and tests are measurable against the codebase and a suite run. This page publishes no revenue, customer, or funding figures.