Reference
Glossary
This site argues in its own vocabulary, and several of these words are used loosely elsewhere in the industry. Here is precisely what each one means when we use it — written to be quotable, including by a machine.
AI-native
- Built so that the AI layer sits underneath every part of the system rather than beside one of them. The test is not whether a product has AI features but whether a question crossing two areas of the business can be answered in one query. AI-added-on demos well and narrows to a single silo in real use. Where this is covered properly
Plataforma de gestión empresarial
- One system covering the work a company runs on — customer records, delivery, support, documents, money and the automation between them — instead of a separate product per function. Broader than a CRM, narrower than an ERP. Where this is covered properly
One data graph
- Every record in every module is a node in the same schema, so a deal, the project delivering it, the tickets raised against it and the invoice for it are genuinely connected rather than matched by a shared customer name. This is the property every other claim on this site depends on. Where this is covered properly
Module
- One functional area — Sales, Projects, Support, Docs, Financials, Marketing, Mail, Drive, Calendar — included in the platform rather than sold separately. Modules share the schema, the permission model and the audit log; they are not applications stitched together. Where this is covered properly
One permission model
- A single definition of who may see and do what, applied to every module and to the AI layer alike. The AI is a caller subject to it rather than an exception to it, which is why an assistant cannot surface a record the person asking could not open themselves. Where this is covered properly
One audit log
- Every action — human or agent — writes to the same trail, with the actor, the authority it acted under, the before-and-after state and a timestamp from one clock. Agent actions are tagged as such, so "show me everything the assistant did to this account last week" is one query rather than a reconstruction across systems. Where this is covered properly
Rollback path
- An AI action goes through the same transactional, soft-deleting write path as a human one, so the inverse of it already exists. There is no second, weaker write route for automated callers — the route that usually gets added under launch pressure. Where this is covered properly
Soft delete
- A deleted record is marked deleted rather than removed, so it disappears from every list while remaining recoverable and auditable. Tenant data is never hard-deleted through the application. Where this is covered properly
Generative AI
- AI that answers and drafts: summaries of a record, replies, explanations of what changed. It reads; it does not act. This is available in the product today. Where this is covered properly
Agentic AI
- AI that takes actions on your behalf rather than only answering — creating, updating and moving records. Actions are where trust is lost, which is why this is the half that needs one permission model, one audit log and one rollback path underneath it. Cyril describes fully autonomous agentic mode as coming soon, not shipped. Where this is covered properly
Operating modes
- Three descriptions of how hands-on the person chooses to be, not of how much AI is present. Platform: the full application, driven by the user. AI: the user asks, and it plans, shows the plan, asks approval and acts where they can watch. Agentic: hand a goal over entirely — coming soon rather than generally available. Where this is covered properly
ai_context serializer
- A deterministic, schema-aware view of a record built for a language model rather than for a database — resolved, related and current, instead of raw rows the model has to decode. Every entity has one, which is what makes a cross-module answer groundable in the same data twice. Where this is covered properly
Grounding
- Tying a model’s answer to specific records it was actually given, so the answer can be checked against them. Thin context is not fixed by a better model: one given a fifth of the data still answers a fifth of the question, just more fluently. Where this is covered properly
Automation engine
- A no-code trigger-condition-action engine that runs at the graph level, so a rule can join modules directly instead of passing records between products. The AI layer calls the same engine, which means an automated action is scriptable and auditable by the same route a human-built rule is. Where this is covered properly
Integration tax
- The cost of running a business across disconnected systems that never arrives as an invoice: the software bought to bridge them, the time spent maintaining the bridges, the reconciliation of several versions of one customer, and AI add-ons that can each see only one silo. On this site’s published model it is about 70% of the total, against roughly 30% for the licences everyone negotiates. Where this is covered properly
Connective software
- The automation or iPaaS products and paid connectors bought to move records between systems that do not share a data model. Necessary in a multi-tool stack and the smallest of the costs it creates — the maintenance of the mappings is larger than the subscription. Where this is covered properly
B2B and B2C mode
- Chosen once at onboarding. B2B is account-and-contact centric; B2C is contact-and-deal centric. The schema, the navigation and the AI prompts follow the choice, rather than showing every team a compromise between the two. Where this is covered properly
Client portal
- The view your customers get of their own tickets, documents, invoices and projects. Portal users are free and unlimited — they are not staff seats, and they do not count towards the per-seat price. Where this is covered properly
Per-seat pricing
- One price per staff seat covering the whole platform — no base platform fee, no per-module upsell, and data export always free. The relevant comparison is not against one tool’s licence but against the total of the stack it replaces. Where this is covered properly
Retention and legal hold
- Rules governing how long files and records are kept, and a hold that prevents deletion while a matter is open. Part of the module rather than a paid tier. Where this is covered properly
Sé de los primeros en usar Cyril.
Únete a la lista de espera para el acceso anticipado. Solo te escribiremos cuando haya algo real que contar.