Which AI providers does it support?

Claude (Anthropic), OpenAI, Google Gemini, Azure OpenAI and OpenAI-compatible servers such as Ollama, LM Studio or vLLM for your own models.

The copilot that does the work. And never more than you can.

Q: Can Bastino see more than I can?

No. Bastino acts as the person asking: each tool checks that person's permissions and screen access. It has no credentials of its own and no administrator mode.

Q: Where does my data stay?

In your instance: each customer has its own database, files and keys. Only what is needed to answer travels to the AI provider, and with your own model inside your network nothing leaves.

Bastino lives on every Fisher Platform screen. Ask it, request it or say it out loud, and it will search, summarize, hand you the Excel, read the contract and make the change. With your permissions, with your confirmation and with your data in your instance.

See it with my data Watch the video

Acts as you, never as an administrator
Nothing changes without your click
You choose the model, even your own

Bastino · Customer file ES

What is missing from the Autopartes Delta file? Send them the link and approve the tax certificate they uploaded today.

documentos_listar file 7 7 pending · 0.2 s

documento_leer Tax status certificate verified citations

7 documents are missing; the mandatory ones:

Requirement	Status
Letter of commission · import	Missing
Program annex I	Missing
Program annex II	Missing

The certificate is dated 10/10/2026 and the RFC matches the record [cite p. 1].

Requirement "Tax status certificate": ~~Delivered~~ → Approved ConfirmDiscard

In one minute

Bastino, working today

Real screenshots of the platform in a demo environment, with fictitious data.

What it can do

It is not a separate chat: it is another member of the team

With access to everything you can see in Fisher Platform, and to the same actions you would take on screen.

Searches everything and summarizes

"What is due this week?", "what do we owe this supplier?", "how are Purchasing and Warehouse related?" It answers with tables linked to each record.

Does what you would do

Issue and send an invoice, record a payment, assign a ticket, rule on a requirement, approve a purchase, move a candidate, record a stage of a procedure… with the same logic as the screen.

Delivers the finished work

Excel, Word and PowerPoint files ready to download in the chat itself, diagrams and tables that copy as they are.

Reads documents and cites them

A contract, a certificate, the PDF you attach: it reads it and every claim carries its citation, checked on the server against the document.

Talk to it

Dictate in Spanish or English and listen to the answer, or talk with it as on a call while you walk the plant.

Works on its own

"Every Monday, send me what is due" or "when an urgent ticket comes in, alert me on WhatsApp": scheduled and event-driven tasks, with your permissions.

Specialists

One expert in collections, another in foreign trade: agents with their own instructions and tools, which can never do more than whoever asks them.

In your portals

Your customer asks about their invoices or their file, your supplier about their payments and your employee about their vacation. Each sees only their own.

Connected by MCP, in both directions

It queries your other systems over MCP, and Claude, ChatGPT or Copilot can query Fisher Platform through Bastino, as the person connecting.

admin/dashboard

Reading is done; changing is proposed. Bastino prepares the change through the same API the platform uses, shows the before and after and waits for your click.

It knows the platform. It answers how something is configured and how the modules relate.

From the data to the Excel, in the chat itself.

Voice in Spanish and English.

Tasks that run on their own, with your permissions.

Specialists with their own instructions.

MCP connections to your other systems.

Secure by design

Powerful must not mean dangerous

An AI that acts inside your ERP has to be, first of all, predictable. This is how Bastino is built.

Your data, in your instance

Each customer has its own database, files and keys. Bastino's conversations, memory and audit live there, not in a separate service.

You choose the model

Claude, OpenAI, Gemini or Azure OpenAI with your own key, or a model of your own (Ollama, LM Studio, vLLM) inside your network: then the conversation never leaves your company.

It is you, not a superuser

Each tool checks your permissions and your screen access. If you can't see it, neither can Bastino. In the portals, each person sees only their own company's data.

Reading is done; changing is proposed

Every change goes through a card with the before and after, in red if it deletes or touches sensitive fields, and is confirmed only once even if you press twice.

Critical actions do not exist

Installing modules, granting itself permissions, lowering security, deleting backups or disabling its own user are out of its reach, and cannot be enabled from any setting.

Against instruction injection

What third parties write —a ticket subject, a document, a response from another system— travels flagged as untrusted data, and links are filtered before being shown.

What Bastino never does

A closed list, closed on purpose

Install, update, activate or deactivate modules: that would be running new code.
Change its own roles or permissions, or grant or remove administrator.
Lower security: second factor, login throttling, passwords, passkeys, single sign-on.
Restore or delete database backups, or delete or deactivate its own user.

It is not a setting that an administrator or an instruction hidden in a document can switch on: what doesn't exist can't be requested. A module can add exclusions, never remove them.

admin/settings?group=bastino_actions

Actions per module. The administrator decides what Bastino can do; the critical ones don't even appear.

MCP server

Claude, ChatGPT or Copilot inside Fisher Platform, with the door firmly shut

OAuth 2.1 with PKCE: the person grants access on a consent screen with their session, and can revoke it.
Tokens are stored as a fingerprint, expire and rotate; they never travel in a cookie.
From outside, a change is never applied: at most it stays proposed in the inbox until its owner confirms it.
Quotas per connection, maximum size per request and an audit of every call.

admin/settings?group=bastino_mcp_server

Governance

It knows what it costs, who used it and what for

Budgets per person, per role, per door and per installation, with an alert when reaching the threshold you set.
Every call to the model measured in tokens and in money, by provider and by model.
Audit of every question and every tool, with the retention you decide, exportable to CSV.
Per-person memory that each person sees and deletes, and a company glossary that you manage.

admin/bastino/usage

Audit of every question.

Governance and memory, configurable.

In honesty: no AI is infallible against malicious text. That is why Bastino doesn't depend on the model "behaving": the critical doesn't exist, what is protected is rejected by the server and everything else goes through your confirmation. The model may make mistakes; your ERP doesn't change without you.

The model you prefer

Without being tied to a provider

It is configured in Settings, per installation, and changed without touching anything else.

Claude

From Anthropic, with your own key.

OpenAI and Azure

OpenAI directly or Azure OpenAI on your own subscription.

Gemini

From Google, with your own key.

Your own model

Ollama, LM Studio or vLLM on your network: nothing leaves your company.

Frequently asked questions

What we are usually asked

Can Bastino see more than I can?

No. It acts as the person asking: each tool checks that person's permissions and screen access. It has no credentials of its own and no administrator mode.

Can it change data without my knowing?

No. Reading is done on its own; any change is proposed on a card with the before and after of each field and is only applied when you press "Confirm". It can also be set to "read-only" for the whole company or per role.

Where does my data stay? Is it used to train models?

Conversations, memory and audit stay in your instance, with its own database. Only what is needed to answer each question travels to the AI provider, using your company's key, so that provider's terms apply; with your own model inside your network nothing leaves.

What if a document contains hidden instructions?

What third parties write travels flagged as data, not as an order. And even if the model got confused, critical actions don't exist and every change goes through your confirmation, in red if it is sensitive.

How much does it cost to use?

Bastino comes included in Fisher Platform. Model usage is paid by each company with its own key, and you can set budgets per person, role or door so it never gets out of hand.

Can I use it from Claude or ChatGPT?

Yes. Bastino is an MCP server with OAuth 2.1: the person authorizes the connection with their session, and from outside it can query with their permissions. Changes are never applied from outside; they stay proposed until confirmed in Fisher Platform.