blueberryEnterprise

An evaluation path for offices, IT teams and laboratories who want to try a Linux desktop with office applications and local inference.

Berry‑AI for IT departments

Berry‑AI can perform textual tasks on the content you provide it. The client returns text. The operator applies changes, executes commands and launches tests with his tools.

Assistance and diagnosis

Paste an error or log excerpt without sensitive data. Ask for an explanation, hypothesis to test, or a sequence of checks for the ticket.

Development and review

Provide a snippet and expected behavior. Ask them to explain the code, propose a targeted change, identify edge cases, or sketch out unit tests.

Procedures and documentation

Turn notes into a how-to, release checklist, or knowledge base outline. Re-read steps and prerequisites before sharing.

An enterprise-level local AI

List of scheduled reports with outcomes, next executions and delivery history.

Reports overview

Check when the next report will start, start one manually and open the delivered file. If a shared folder is unavailable, the history shows which execution failed and why.

Module with SQLite source, weekly schedule and XLSX delivery to a shared folder.

Configuring a report

Choose read-only database, test the query and set days and times. The report can be generated in XLSX or PDF and delivered via email or to a shared folder. Summarization with Berry‑AI is optional.

Qwen 1.5B in use, 7B recommended and RAM warning for 14B.

Catalog of local models

The catalog combines the pre-installed Qwen2.5 1.5B model with the 7B and 14B variants, which can be downloaded for free with the Apache 2.0 license. The proposal evaluates RAM and free space to suggest which model to install and use offline.

Request for project documents with preview of the draft and button to confirm sending.

Agent Mode a use case

The agent opens Thunderbird, searches for the recipient and prepares an email to request the necessary project documents from the customer. Before sending, it shows the recipient, subject and text, with the possibility of modifying the draft or canceling. The address shown is fictitious.

Create Agent screen with the Write Code action, instructions for reading CSV files, and a text editor selected.

Create agent

Save an agent with a name, action type, and reusable instructions. In the example, Python Assistant generates code to read CSV files and check inputs. You can choose an app to open after confirmation, such as a text editor to review the result.

Python Assistant shows the request to read a CSV, the generated code and the confirmation to open the text editor.

Python assistant in action

Ask the agent to write a function that reads a CSV file and returns the rows as dictionaries. Berry‑AI shows the code to review and copy. The text editor opens when you choose Confirm and open.

From the VM to the first workstation

  1. Check your work every day

    Test network, audio, printing and screens on the designated locations. Open representative documents, templates and spreadsheets in LibreOffice. Also check shared mail, calendars, and streams.

  2. Define rules for local AI

    Decide what data goes into prompts, who approves models, and how results are shared. Use non-sensitive examples in the pilot, compare quality and times on team cars.

  3. Prepare a replicable configuration

    Record model, version, applications, updates and test results. Define who handles backup and recovery before moving to other locations.

What to evaluate with Boreale

The pilot starts from Boreale's ISO AMD64. Centralized management services, certifications and SLA assistance remain separate areas of evaluation.

Writer in Night on a Boreale station.

Information on the site

The desktop simulation uses Boreale screens. The Berry‑AI examples have predefined responses. The theme preference remains in the browser.