Memory Cross-conversation memory is on by default — but it's implicit synthesis, stored on OpenAI's servers. You can't open it as structured data, query it, or take it with you. A structured knowledge graph — entities, relationships, history — in a database you own. Entity-aware memory that tracks your business and recalls it across threads today; a richer subject-graph mode (dedup, project hierarchy) is rolling out.
Business context It picks up your preferences and style. There's no structured record of clients, deals and numbers you can browse and check. Tracks the entities of your business — clients, deals, KPIs — as structured records you can browse and search in the graph view.
Model choice One vendor. OpenAI's models, on OpenAI's pricing and deprecation schedule — the product and the model are the same company. The model is a setting, not a foundation. Anthropic and Mistral native; custom or local models via an OpenAI-compatible adapter. Switch providers without migrating your data or workflows.
Actions Connectors and MCP now reach a real set of apps — running inside OpenAI's product, on their infrastructure. Captures contacts and deals, creates tasks, runs workflows, sends email through your own Gmail or SMTP — on software you own.
Background work Scheduled tasks exist (capped per plan) and can call connectors — in their cloud, detached from a datastore you own. A 24/7 background worker bound to your own datastore — cron, page-watch and manual triggers, budget caps, every run reported back to you.
Integrations First-party connectors plus custom MCP servers — a growing but curated set, vendor-hosted. Connect-any-API: point it at the API docs and the agent builds a typed, reusable integration — reaching the long tail no connector catalog covers.
Data ownership The vendor hosts your data; consumer plans run on US infrastructure (EU data residency is an Enterprise option). Self-host or managed. Your data lives in a database you can copy, on infrastructure you control — EU data centers available. The source is open (ELv2) to read and fork.
Cost Per-seat subscription tiers, priced and metered by OpenAI — see their current pricing. Self-hosted: free, plus your own API usage. Hosted from CHF 39/mo. All features on every tier — no per-seat, no per-task metering.
