Skip to content

KastorLab sets up secure AI in your company: a dedicated instance in Europe, or on your premises. Audit me →

Nestor · KastorLab's AI toolbox

The right model for each task, in your dedicated instance.

Nestor is the orchestrator that runs AI in your company, on an instance dedicated to it. It sends each task to the right model, so everyday work doesn't pay premium-model prices. Your agents run isolated from the network.

Nestor's home screen: an "Ask Nestor" conversation box, the Ask and Build modes, the Auto model picker, and four shortcuts: summarise a document, prepare a follow-up, create a memory, run an agent.
Real screenshot of Nestor, version 0.1 (French interface).

For your line of work

Your company's repetitive tasks, handed to Nestor.

  • Accounting

    Entries, VAT checks, bank reconciliations.

  • Finance

    Cash flow, forecasts, unpaid-invoice reminders.

  • Audit

    Ledger tests, analytical review, confirmations.

  • HR and payroll

    DSN checks, payroll variables, termination pay.

  • Legal

    Billable time, conflicts of interest, contract review.

  • Compliance and risk

    KYC files, watch lists, recurring checks.

  • Sales

    Account scoring, follow-ups, draft letters.

  • IT and AI governance

    Track and reduce token spend.

Cost per task

The right model for each task.

In automatic mode, Nestor first picks the least expensive model capable of the task. A model installed on your own infrastructure isn't billed per token.

  • Open-weight models served locally (Qwen, Gemma, Granite, Mistral, gpt-oss families).
  • An external model only if you allow it: every outbound request is logged and flagged in the conversation.
  • A Costs screen estimates what your team spends.

Ready-to-use agents

Agents that run with no model.

12 included business templates (accounting, legal, compliance, sales) run with no AI model. For a need specific to your company, AI writes the agent from the structure of your files, never your actual rows.

  • A program agent run calls no model: 0 tokens.
  • Tested on your files, cell by cell, up to 3 attempts.
  • Daily, weekly or monthly schedules, in your team's time zone.
  • Before creating, Nestor reuses an existing agent: “check this morning's invoices” runs it, 0 tokens, and the answer says so.
  • Connected to your tools: Atlassian (Jira, Confluence) and email, each person with their own rights. Other services (Google Drive, Gmail, Outlook) are supported but not yet tried on the real service.

Security

Secure AI, in a closed loop.

Nestor runs on an instance dedicated to your company, in Europe or on your own infrastructure. Agents run in a closed loop: in an isolated, read-only container, with no network access.

  • "Internal only" mode: no external model, no web search, all the AI stays in your instance.
  • Automatic masking of some personal data (emails, IBANs, numbers…) before any text is sent.
  • The model that writes an agent receives the structure of your files, never your actual rows.

Human approval

Human approval for every agent.

The author of an agent or a change can't approve it themselves. Everything is versioned, traceable and exportable: a basis for preparing your compliance (GDPR, AI Act).

  • Version history; a change is replayed on your files and on every past period kept, with a summary for the approver.
  • When a file changes format, the run stops, explains the difference and proposes a repair for approval.
  • Your team corrects the wrong rows of a result: Nestor infers the kind of mistake and proposes a fix, subject to approval.
  • Audit pack per agent and per period (versions, approvals, fingerprints of files read), sealed for integrity.
  • Quarantine: after several consecutive failures, automatic runs are suspended.

Demonstration

Five minutes, on your own file.

  1. Describe the task in a sentence and attach your file.

  2. See what is sent to the model: the structure of the file and made-up rows, never your real rows.

  3. Watch the test on your file, cell by cell, against your team's result.

  4. A colleague reviews and approves the agent.

  5. The agent runs: 0 tokens, €0 of model.

Book a demo

Deployment

An instance that fits you, with us.

KastorLab sets up your instance, trains your teams and runs Nestor for you: in Europe, or on your infrastructure with the Sovereign offer.

Nestor is in active development (version 0.1). Access is opening gradually.

  • Team and Business offers: a dedicated instance run by KastorLab in the European Union, nothing to install on your side.
  • On your premises (Sovereign offer): Linux server (Debian 12 or Ubuntu 24.04) or workstation (macOS, Linux, Windows via WSL 2 with Docker Desktop).
  • Docker required: it isolates agent runs.
  • Local models: a graphics card with at least 12 GB of video memory (Apple Silicon Mac: 16 GB of unified memory or more).
  • Program agents that are already built need no model at all.

Frequently asked questions

Nestor in questions.

What is Nestor?

Nestor is KastorLab's AI toolbox. It turns a task described in plain language into a program agent: AI writes the agent from the structure of your files, never from your actual rows; the agent is tested on your files, approved by someone other than its author, then run with no model, at 0 tokens per run.

What is a program agent?

A program written by AI, versioned and reviewed, which then runs without calling any model: in an isolated container, with no network access and a read-only file system. It can run on a schedule (daily, weekly, monthly). Conversational agents and the chat, on the other hand, call a model at every turn.

How is Nestor different from an online AI assistant?

An online assistant answers each request by calling a model. Nestor turns repetitive work into a program that then runs with no model: tested on your files, versioned, approved by a colleague and replayed on past periods before each change. When a file changes format, the run stops and a repair is proposed, for approval.

Is Nestor open source?

No. Nestor is proprietary licensed software, installed for each company on an instance dedicated to it. It relies on open-weight models (Qwen, Gemma, Granite, Mistral, gpt-oss families) and, if your organisation allows it, on external models (Claude, GPT, Gemini).

Can Nestor work without any external model?

Yes, in “internal only” mode: no external model, no web search. Agents are then built with a local model, which is slower and may not be enough for complex tasks. Agents already built need no model at all.

Which tools does Nestor connect to?

Atlassian (Jira, Confluence) and email (IMAP, SMTP) have been tried on the real services; each person signs in with their own rights, and an email is only sent after a click. Google Drive, Gmail and Outlook are supported, but not yet tried on the real service.

Where does Nestor stand today?

Nestor is in active development (version 0.1). Access is opening gradually: book a demo to see it on your own files.

See Nestor on your company's tasks.

Book a demo, or request an audit of how your company uses AI: we start from your real tasks.

Request an audit

Tell us who you are: we'll get back to you to scope an audit of how your company uses AI.

Book a demo

Pick a slot: you receive an invitation with a Google Meet video link.