AI workspace, installed on your infrastructure

Ask your documents, email and chats. Get answers that cite the source.

Nexa, the AI assistant of Mislata, searches the files, mailboxes and Slack or Teams conversations your teams connect, answers with citations, writes documents from your templates and uses the tools you allow. Mislata is installed in your own cloud or data centre, on Kubernetes or a single server. Documents, chats and indexes stay there, and prompts go only to the model endpoints you approve.

Request a deployment reviewSee what it does

What it does

Six things people do with Mislata.

Each one works inside a Space, under the same access rules and the same policies.

  1. 01

    Ask across your sources

    Ask Nexa, and it answers from the documents, email and Slack or Teams messages indexed in a Space. Every answer lists the passages it used, and each citation opens the original.

    For exampleWhat did we agree with the client about the delivery date?

  2. 02

    Organise work in Spaces

    A Space per client, matter or project, with its own sources, members and roles. Guests can be invited with an expiry date.

    For exampleCreate a Space for the new office lease and invite the external counsel.

  3. 03

    Use tools in the chat

    Nexa can search the web through the search service you run and call the tools of the MCP servers your administrators register.

    For exampleCheck this week's exchange rate and update the cost table.

  4. 04

    Write documents from your templates

    The document generator drafts long documents from your organisation's templates, lets you revise them with an assistant and exports to Word, PowerPoint, PDF or Markdown.

    For exampleDraft a twelve-month services agreement from our standard template.

  5. 05

    Reuse prompts, share results

    Keep prompts in a shared library. Export an answer to Word, PDF or Markdown (Excel when it has a table), or send it to a Slack or Teams channel after the policy check.

    For exampleSend this summary to the project channel.

  6. 06

    Apply policies, keep a record

    Before each model call, policies can redact personal data, block forbidden words and enforce token budgets. Every call is recorded with user, model, tokens and cost.

    For exampleRedact personal data in every prompt before it reaches the model.

In the product

The product as it is today.

Screens from the current version, in a demo tenant set up as a law firm. The number next to each capture refers to the list above.

Demo tenant of a Spanish law firm: the data is in Spanish and fictional.

Chat with Nexa in Mislata: a question about a services contract, an answer listing the clauses to review, and a cited fragment of the uploaded contract.
01A cited answer. The question is about a contract uploaded to the Space. The answer lists the clauses to review; each source shows the fragment it used and opens the file.
Document generator: eight legal templates, including a bilateral NDA, a services contract and a GDPR privacy policy.
04Document templates. The firm's templates, grouped by category, are the starting point for the document generator.
Compliance page: compliance score, counters for audit events and policy results, and a checklist.
06Compliance view. An overview for the EU AI Act and GDPR: AI systems in use, audit events, policy results and a checklist, with the audit log exportable as CSV.

Organisational context

One question across the systems where work happened.

A decision rarely lives in one file. Nexa answers from the approved material in a Space: documents, selected email and selected conversations. Each answer cites the original item. Space membership, source ownership, document access lists and ethical walls decide what may be retrieved.

Example question

What changed in the proposed contract, who approved the exception, and what was agreed with the client?

  1. 01 / Documents

    SharePoint, OneDrive, Google Drive and uploaded files

    Indexed into a Space. Answers cite the file.

  2. 02 / Email

    Outlook and Gmail

    Messages and supported attachments from mailboxes a user connects. Mislata does not read every mailbox.

  3. 03 / Conversations

    Slack and Microsoft Teams

    Messages a user chooses to index into a Space. Citations link back to the original message where possible.

Information boundaries

Breadth, inside a boundary.

More sources only help if each answer stays where it belongs. Four controls decide what retrieval may return.

  1. 01

    Teams and Spaces

    Work is grouped in Spaces with Owner, Editor, Viewer and External roles. Tenant teams share chats, knowledge and documents.

  2. 02

    Document access lists

    An access list on a document restricts who can retrieve it. Retrieval skips what the user may not read.

  3. 03

    Connector ownership

    Email and messages indexed from a user’s connection stay owned by that user in retrieval.

  4. 04

    Ethical walls

    A wall between two Spaces stops anyone from being a member of both and stops Shared-Space retrieval from crossing it. It is not a complete conflict-checking system.

Where it runs

In your environment, with your models.

Mislata installs in your Azure subscription, a Kubernetes cluster or a single server. Identity, data, files and secrets stay in your environment, and you choose which model endpoints it may call.

How it deploys

Your environment

Azure subscription, Kubernetes cluster or a single server

Mislata
Web app, API, document generation and export
Identity
Microsoft Entra ID or QuantumID
Data
PostgreSQL with pgvector, Redis
Files
S3-compatible storage or Azure Blob
Secrets
Kubernetes Secrets or Azure Key Vault

Model endpoints you approve

Azure OpenAI and Foundry, Anthropic, OpenAI-compatible

Sources you connect

SharePoint, OneDrive, Google Drive, Outlook, Gmail, Slack, Microsoft Teams

Deployment boundary. Each model endpoint can run in your subscription or outside it; you choose which ones Mislata may call.

Where to start

The first question each buyer asks.

Next step

Start with a deployment review.

Tell us where Mislata should run, which identity provider you use and which models you have approved. We go through the architecture with your team before anything is installed.

Request a deployment reviewOr write to hello@mislata.eu