iVendNext Copilot is a conversational AI assistant embedded directly inside iVendNext Desk. It lets you ask questions in plain language, trigger actions across modules, and retrieve data from your live iVendNext instance — all from a persistent chat widget that sits alongside your normal work interface. There is nothing to install, no separate application to log into, and no data sent to an external platform without your own credentials authorising it.
This article explains what Copilot is, how it fits into iVendNext, the BYOK (Bring Your Own Key) model that governs its AI connection, and what it is and is not designed to do.
Copilot is embedded in iVendNext Desk as a persistent chat widget. It is always available in the interface — you do not need to navigate to a separate screen or module to use it. You type a question or instruction, and Copilot responds in the same panel.
This placement is deliberate. The goal is to reduce context-switching: instead of opening a report, running a query, navigating to a DocType, or calling a colleague for a data point, you ask Copilot directly and get an answer without leaving what you were doing.
Copilot is not a standalone SaaS product, a mobile app, or an integration bolt-on. It is part of iVendNext Desk, delivered as a feature of the platform.
Copilot does not come with a built-in AI provider. Instead, it uses a Bring Your Own Key (BYOK) model — you connect your own AI account, and Copilot uses that connection to process your requests.
Two providers are supported:
ChatGPT (OpenAI) — connected via OAuth PKCE. You authorise the connection through your OpenAI account; no API key is stored in iVendNext.
Anthropic Claude — connected via API key. You generate a key in your Anthropic console and enter it in iVendNext settings.
This model has a specific implication: iVendNext does not hold, proxy, or intermediate your API key or your conversation data. The conversation between Copilot and the AI model goes through your own provider account. Your data relationship is with your chosen AI provider, governed by their terms of service — not by a third party sitting between you and the model.
BYOK also means your organisation controls the AI account. You can revoke access, rotate keys, monitor usage, and apply your own provider-level controls at any time.
Copilot has two core modes of operation.
Tool execution. Copilot has access to 40+ pre-built tools that map to specific iVendNext functions across Stock, Sales, Purchasing, CRM, and core accounting operations. When you ask it to do something — create a draft Sales Order, look up a customer's outstanding balance, check stock levels for an item — it identifies the right tool, calls it, and returns the result. Every tool call is gated against your iVendNext permissions; Copilot cannot execute an action your account is not authorised to perform.
Data retrieval. For questions that require pulling data from multiple tables — sales by region, items below reorder level, overdue invoices by customer — Copilot constructs and executes a query against your iVendNext database in plain language. You do not need to know which tables to join or how to write SQL. You ask the question; Copilot handles the query.

Both modes operate on your live iVendNext data. Copilot does not work from a snapshot or a cached copy.
It helps to be equally clear about what Copilot does not do.
It is not an autonomous agent. Copilot does not monitor your system, take initiative, or run in the background. It responds to what you ask, within the scope of a conversation. There is no scheduled task execution or background automation — every action starts with something you type.
It does not submit documents automatically. When Copilot creates a Sales Order or Purchase Order, it creates a draft. Your team reviews and submits. Copilot does not post, submit, or finalise any document without human review.
It does not have access to everything. Each DocType in iVendNext has a field whitelist — only the approved fields are visible to Copilot. Sensitive fields such as passwords and API keys are blocked. Certain DocTypes (User, Role, System Settings) are entirely off-limits. Copilot works within these boundaries by design.
It is not a recommendation engine. Copilot retrieves and presents information. It does not tell your team what to buy, who to call, or what decisions to make. The judgement stays with your people.
Every conversation with Copilot is primed with a snapshot of your current iVendNext environment. Before your first message is processed, Copilot receives:
Your company snapshot (name, base currency, active fiscal year)
Top customers by recent transaction volume
Key items flagged for attention
Overdue alerts from across the system
This context means Copilot's responses are grounded in your data from the start of every session. You do not need to tell it which company you are working in or provide background on your business — it already has the current state.

Copilot responses stream token-by-token via Socket.IO as the AI processes your request. You see the answer appearing as it is generated — you do not wait for a complete response before reading it. For longer answers or data queries, this makes the experience noticeably faster and more interactive than a batch-response model.
Because Copilot uses your own AI provider account, the data handling policies that apply are those of your chosen provider (OpenAI or Anthropic), not a separate iVendNext data policy for AI. Before deploying Copilot, your organisation should review your chosen provider's data processing terms and ensure they are compatible with your own data governance requirements — particularly if your iVendNext instance holds personal customer data or commercially sensitive records.
