---
id: KB-MA-001
url: https://app.codecontract.io/help/marta/what-is-marta
idioma: en
categoria: marta
audiencia: usuario
actualizado: 2026-08-13
tambienEn: [es]
relacionados: [KB-DI-001, KB-PS-001, KB-TZ-001]
citadoPor: [KB-MA-002]
enLaApp: https://app.codecontract.io/marta
---

# Marta, the AI assistant

_What she knows about your organisation, what she can do for you and where the limit is._

**Responde a:** what is marta · the platform's ai assistant · can I ask the ai about my cases · what does marta know about my organisation · can the ai do things for me

Marta is an assistant that works on what is already in your organisation: your cases, your documents, your contacts. The difference from a generic assistant is that she does not answer from memory: she answers by looking at your data, and only the part you are allowed to see.

**En corto**

- She sees only what your user sees: no permission jumping, no crossing organisations.
- She is for asking and for doing, not only for chatting.
- What she does is logged exactly as if you had done it yourself.
- She does not decide anything with consequences: she proposes and you confirm.

## What she is actually for

- Finding something without remembering where it was: "the contract we signed with Gómez Workshops".
- Summarising status: "which cases have been stuck for more than a week".
- Preparing work: describe a process in plain language and she builds it for you to review.
- Answering how-to questions without leaving the platform.

## What she does not do

She does not see what your user cannot see, nor data from another organisation. She does not sign for you or close a case on her own. And she does not invent: if something is not in your data, she says so rather than filling it in.

> [!NOTE]
> Actions she takes appear in the log with a record that they went through her, just like any other. Traceability does not change because you used the assistant.

## How to ask well

| Instead of | Try |
| --- | --- |
| "find me the contract" | "the rental contract we signed with Ruiz in March" |
| "how is everything going" | "which supplier onboarding cases are incomplete" |
| "make a process" | "a process to ask suppliers for their tax ID, insurance and clearance certificate" |

_The more specific the question, the fewer rounds — exactly as when asking a person._

## Frequently asked questions

**Can she see another department's documents?**

Only if your user can. Permissions are the same as browsing by hand.

**Are models trained on my documents?**

Your organisation's data is used to answer you. If you need the contractual detail of how it is processed, it is in the legal documentation and the trust centre.

**Does asking her consume credits?**

AI queries have a cost, like automatic reading. It shows up in the consumption history.

**Can she be wrong?**

Yes, like any assistant. That is why what she proposes gets reviewed before you confirm, exactly like a field read from a document.

## Ejemplos

**Someone needs to know which supplier onboardings have been open more than two weeks and does not know how to filter.**

- Asks Marta directly, in those words
- Gets the list with what is missing on each
- Asks her to resend the reminder to those who have not delivered

→ Ten minutes of filtering turned into one question, and the reminder goes out with a record of who asked for it.

**A manual search runs across twenty files.**

- Asks and lets it assemble the answer

→ The search goes from a morning to a minute.

**Something is asked that the answer cannot know.**

- Checks how far what it sees extends

→ Expectations are set beforehand.

**The answer is taken as final.**

- Checks which documents it comes from

→ The answer can be verified.

**The question is asked in an ambiguous phrase.**

- Narrows it to the file or the period

→ The answer points at what was sought.

**It is used to draft something that goes outside.**

- Reviews the text before sending

→ What goes out carries your judgement.