Private and open-source AI

AI that stays inside your boundary.

Some workloads cannot leave your network. We deploy open-source models you can run, govern and own, in your cloud account, on your servers, or fully offline.

Right-sized models, boring supportable stacks, and an exit path if your needs change.

Plan my private AI Where should it run? ↓
  • Your boundaryCloud account, servers or offline
  • Open-sourceModels you can swap and govern
  • RunnableMonitoring and update discipline
where should the AI run? · illustrative

How sensitive is the data the AI will use?

    Illustrative. Real recommendations depend on your data, rules and workload.

    How we compare

    Public API, DIY, or private AI you can actually run.

    Public APIs are fine until compliance, latency or cost say otherwise. Then you need a plan for models you run yourself.

    Where should it run?

    Start from the rule that constrains you.

    Pick what sounds like your situation. We match model size and hosting to the job.

      First slice:

      Start from here

      What we build

      Four things that keep private AI runnable.

      Private inference platforms, open-source LLM deployments and secure environments for local or VPC-hosted AI.

      • Model choice and sizing

        The smallest model that does the job well.

      • Secure hosting

        VPC, your servers, or fully offline.

      • Access control and monitoring

        Who uses it, and what it is doing.

      • Update discipline

        Models and dependencies updated without surprises.

      What good looks like

      AI your team can trust, and your auditors can inspect.

      The goals we plan around, agreed with you at the start.

      • Data stays inside

        Your boundary, your rules.

      • Governed and auditable

        Access and usage you can review.

      • Runnable by your team

        Boring, supportable and documented.

      • An exit path

        No dead end if needs change.

      Outcomes depend on your situation. We set goals and measures with you before work begins.

      How we work

      From boundary to a running assistant.

      Match model size and hosting to the job, prefer boring stacks over novelty, and keep an exit path.

      1. 01

        Define the boundary

        Where data must stay, and who may use it.

        You get: clear constraints
      2. 02

        Choose model and hosting

        Right-sized, supportable, and ownable.

        You get: a sound design
      3. 03

        Deploy securely

        Access control and logging from day one.

        You get: a protected deployment
      4. 04

        Operate and monitor

        Health, usage and quality watched.

        You get: confidence
      5. 05

        Update and evolve

        Models and dependencies updated with discipline.

        You get: a system that stays current

      Proof

      Principles we hold ourselves to.

      Published work and our principles. Measured client results are shared only with the client's permission.

      How we run private AI so it stays runnable

      Right-sized
      We match model size and hosting to the job.
      Supportable
      We prefer boring, supportable stacks over novelty.
      No lock-in
      We keep an exit path if your requirements change.

      Partner program

      Your client cannot use public AI. We build the private alternative with you.

      Agencies and consultancies bring us in to deliver this under their own brand, or alongside their team. Scope, ownership and responsibilities are agreed in writing first.

      • Your client relationship
      • Your brand and pricing
      • Your project governance
      • Your choice of what is disclosed
      Become a partner
      • White-label deliveryWe deliver it under your brand.
      • Co-deliveryOur engineers inside your team and process.
      • ReferralIntroduce a client; terms agreed in writing first.

      Questions

      Before you decide.

      Which models do you use?

      Open-source language models, chosen and sized for your job. We explain the trade-offs and keep an exit path if your requirements change.

      Where can it run?

      In your cloud account (VPC), on your own servers, or fully offline on site, depending on your constraints.

      Is private AI as good as public APIs?

      For many business tasks, right-sized open models perform well, especially when grounded in your own documents. We test on your real use case and tell you honestly.

      Do we need GPUs?

      It depends on the model and usage. We size hosting to your real workload and say plainly when a smaller model is enough.

      How do you keep it maintained?

      Monitoring, access control and update discipline are part of the design, so private AI stays runnable.

      Can it connect to our software?

      Yes. We connect private AI to your documents, workflows and applications, including our business software and automation services.

      Plan my private AI

      Tell us what must stay inside.

      A short description is enough. We reply by email with fit and a sensible next step. No commitment.

      • Your boundary: cloud account, servers or offline
      • Open-source models you can swap
      • Monitoring and update discipline
      • An exit path if needs change

      We use these details only to reply. See our privacy policy or use the contact page.