Notes
Short, first-hand notes on how Historis works and the choices behind it — the data model, the MCP tools, why an event-based CRM behaves differently from a task list, and how it compares to the tools you already use. Written for the people connecting their AI to it.
Track customers without a spreadsheet
Your customer spreadsheet goes stale because updating it is a chore. Here is what keeping track actually requires, and how to move off the grid for good.
Is an MCP CRM too technical for you?
You do not need to be a developer to connect your AI to an MCP CRM. It is one address you paste once. Here is what setup involves and what changes day one.
Customer record: what to note, how to keep it
What a shop's customer record should contain, the 30-second note that keeps it alive, what GDPR lets you write, and the CNIL three-year retention rule.
GDPR customer records: the CNIL rules for shops
What GDPR really requires of a shop's customer file: legal bases, CNIL retention periods, banned entries, the register, and the EUR 2.25M Carrefour fine.
Customer retention: memory beats the points card
Consumers hold 17.4 loyalty programs each; recognition is what retains. The real sources behind the retention stats, and the client history a shop can keep.
Todoist vs Historis: why your tasks need people
Todoist tracks tasks; it has no idea who asked for them. Historis links every task to a person and their history. A factual, feature-by-feature comparison.
One client, one record: email-first resolution
AI assistants split contacts because names vary between captures. Historis resolves people by email first: one client, one record, enforced by the database.
Event-based CRM: events, not tasks, one timeline
An event-based CRM records what happened rather than what you plan to do. Your team and its AI assistants write to one shared timeline, the company's memory.
Bring your own AI: your CRM runs no model
Historis runs no model server-side. You connect your own AI over MCP, even a self-hosted open-source model, so client data never reaches a third-party AI.
How AI-agent search works over an MCP CRM
What your AI actually gets: deterministic Postgres full-text search, scoped to the records it may see before matching, and consistent the instant you write.
Tenant isolation: a leak-proof AI agent surface
Your AI holds a token; tenant isolation means it can only ever reach your data. The architecture that makes a cross-tenant leak fail CI, not slip review.
Why an AI writing to your CRM is safe by design
An AI writing to your CRM is safe only if every write is idempotent, attributable, and bounded, edits are reversible, and deletes are explicit and logged.
Automation rules with no server-side AI
Historis automation rules are markdown you write; the server matches them with exact, reproducible logic, and your own AI interprets and applies them.