What the assistants can see and do

An assistant runs as you. It can look at the same records you can open — no more — and every tool it uses is one we built inside ALFA OS, subject to the same access rules and audit trail as a person clicking the same button. A portal customer’s assistant sees only that customer’s data. Assistants do not browse the open internet on your behalf unless a specific research tool says so when it runs.

Assistants draft, summarise, search and suggest. Where an assistant is allowed to change a record — post an accounting entry, send a message, create a document — it does so under your name, the action is logged like any other, and anything that matters (money, customer-facing messages, ledger changes) either asks you first or follows the same rules a person must follow: one journal, no lumped or offsetting entries, every entry with its party. You are responsible for reviewing what an assistant produces before you rely on it.

The text you type, the audio you speak, and the records the assistant reads to answer you are sent to third-party AI model providers to generate the response: Anthropic and OpenAI for language models, and ElevenLabs for converting voice to text and text to voice. Each provider processes the content on our instructions under its business terms. Where the provider offers it, we use the setting that excludes our data from being used to train their models. We keep conversations for up to 12 months for support and quality, then delete them, as the Privacy Policy describes.

Assistants can be wrong, and they can be confidently wrong. Do not paste passwords, card numbers, or anything you would not put in an email into an assistant. Formulas and pricing you discuss with an assistant are covered by the confidentiality section of the Terms of Use. If an assistant does something it should not have, tell us at info@alfamfg.com and we will look at the log.