Vorkar is a diagnosis layer, not another silo. It reads your systems where they already live, relates the same record across all of them, enforces what may reach the model, and answers in plain English — always with its sources.
SO-4471 stopped between Salesforce and the billing app — the API credential expired at 02:14. 3 orders held.
Sources: Salesforce · Legacy billing · sync.log
SO-4471A customer's order was created in the CRM but never showed up in the ERP. Nobody knows why. Four teams, four systems, four tabs open. Here's how Vorkar answers it in one question.
You don't install a "Salesforce connector" and a "SAP connector" and wait for a vendor to ship the next one. You connect by protocol — SOAP, REST, SQL, NoSQL, document stores, local files, chat. Point Vorkar at an endpoint, give it credentials, and its schema is discovered automatically. The system appears on the canvas as a node orbiting Aura, with its live status, type badge (record / log / infra / knowledge) and security badge (OAuth 2.0, API key, DB user).
The same business object has a different name and a different key in every system. An Opportunity in the CRM is a Sales Order in the ERP is an order_id in the logs. The entity map declares those relationships once, so Vorkar recognizes one real-world order across every system it touches — and can join related records on demand.
po_no = SO-4471), pulls every sync-log event carrying that order_id, and checks the infra health windows around those timestamps. Four systems, one timeline.Before anything is sent to Aura, the record filter decides what's allowed. You author rules visually — grouped AND / OR / nested conditions over each system's real fields — at three explicit layers: per system+object ("from Salesforce Opportunities where Amount > 10,000"), per entity ("this customer, in every system it appears in"), or per field ("never send salary or cost, anywhere"). Matching data still flows through your pipeline and is still counted — only the model is blocked.
Ask in plain English: a specific record, a broad "what's happening today?", or anything in between. Aura answers only from what the connectors actually returned, cites every system it queried, and lets you drill from the answer down to the raw record in the source system. If the data doesn't support a conclusion, it says so. A quiet day gets an honest quiet answer — never an invented summary.
The only write path is propose → approve → execute → undo. Vorkar can rebuild those 13 failed orders — but only after a human approves, and every step is audited.
Build once against Dev or UAT; deploy the same thing to Production. Credentials and environment-specific values are supplied in Extensions — production secrets never live on the canvas.
Every run is listed with its records in/out and status. Paste any id to trace its full parent → child → grandchild journey across systems.