The assistant

The assistant answers questions about the firm's work in plain words and proposes the actions it is asked for. It reads only what the person asking may read, through the same checks their own screens make: a matter behind an ethical wall they are not inside is never named to them. Nothing changes until a person approves it.

Turning it on

The assistant is off until the firm turns it on. Under Settings, Communications, Assistant, Settings, the managing partner switches Assistant on and chooses which models the firm allows, the model and effort a new conversation starts with, and the highest effort anybody may choose. The firm receives its welcome credits the first time it is used.

Turning it on sends what people ask, and the records the assistant reads to answer, to Anthropic for processing. Nothing is kept by the provider after the answer, except for a model the firm opts in to with 30-day retention. The firm records its lawful basis for this under the Data Protection Act, 2012 (Act 843) when it turns the assistant on.

Asking

The Assistant button in the top bar opens the panel beside the page, so the work stays where it was. Typing "AI", "chat" or "assistant" into the search (Ctrl+K) opens it too. On a desktop, drag the panel's left edge to make it wider or narrower; it keeps the width you leave it at.

New chat begins a conversation at once, on the firm's default model and effort; begun on a matter's page, the conversation is about that matter. A conversation with nothing said in it yet is already new, so New chat stays on it. History opens your earlier conversations, newest first; pick one to go back to it, or Back to return to the chat you were in. The X beside a conversation deletes it, after you confirm: it leaves your history, and whoever manages the assistant can restore it from the firm's deleted records.

Press Enter to send, Shift and Enter for a new line, and Stop to end an answer early. Answers are formatted: headings, lists, bold and code read as they should.

The assistant can:

• answer questions about matters, parties, tasks, deadlines and the time recorded;
• record time on a matter;
• add a task or a deadline;
• draft a document from one of the firm's templates;
• open a matter, adding the parties and running the conflict check;
• answer questions about Docavra itself, from this help centre and Docavra's legal and pricing pages, with a link to the page it used;
• open a support ticket with Docavra, when the help does not answer.

Asked how a screen works, about the plans and credits, the terms or how the firm's data is kept, it answers only from Docavra's own pages. When they do not cover the question, it says so and offers to open a ticket; it never makes up a feature, a price or a policy. It speaks for Docavra, so it does not recommend or review other products.

Saying whether an answer helped

Under each answer, Helpful and Not helpful tell Docavra how it did. Not helpful asks what was wrong; the note is optional. The verdict and the note go to Docavra to improve the assistant, so leave clients' names and details out of the note. The conversation itself stays with the firm.

Approving what it proposes

Each action arrives as a card with the same fields its screen has. Change anything that is wrong, then Approve; the card says what became of it and opens what it made. Set aside leaves everything as it was. An action the firm's records refuse (a closed matter, a missing activity code) is marked refused with the reason, and nothing is changed. A card is approved once, however many times it is pressed.

Models and effort

Under the message box, the conversation's model and effort each open a menu to change them; the change applies from the next message. Each model the firm allows is listed. Effort is how hard the model thinks before answering, from low (quick answers, fewest credits) up to the highest the firm allows: higher effort gives a more careful answer and costs more credits. Some models are kept by their provider for 30 days, and stay closed until the firm accepts that under Settings, Communications, Assistant, Settings, Models that keep what is sent, Accept for the firm. The acceptance is recorded with who gave it and when.

Credits

A credit is one pesewa. Each answer costs what its model used, at the price for that model, and shows its credits under it. Credits left and Spent this month are on the assistant's settings page; Buy credits opens Settings, Firm, Subscription, Add-ons, where packs are bought on any plan. Docavra may set a monthly ceiling for the firm; the settings page shows it beside the month's spend.

Limits for each person

A person's monthly limit is the first of these that is set:

1. their own limit, set on Use by person;
2. the limit for their role, under the assistant's settings (the most generous of their roles);
3. the firm's monthly limit.

Nought means no limit. Use by person lists what each person spent in a month against the limit that applies to them, those over it first, and Set their limit on a row gives a person a limit of their own, with a reason; emptied, their role's or the firm's limit applies again. Every change is kept in the audit trail.