● For compliance-led teams

Self-hosted AI assistants for organisations that can’t send their data to a public chatbot

Open Assistant runs on your infrastructure — your servers, your EU data residency, your compliance team’s approval. We help you deploy it, integrate it with the tools your team already uses, and operate it long-term.

Book a 30-minute architecture review → See it running on your stack →

Source-available · BSL 1.1 · 88+ tools · FastAPI · no per-execution fees

The constraint

If a compliance officer, DPO or CISO has ever told you “we can’t use ChatGPT for that” — and you’ve spent the last year routing around that constraint with worse tools — we’re the team that fixes the constraint.

Open Assistant is a source-available AI assistant you deploy on your own infrastructure. 88+ tools, BSL 1.1 licence, no per-execution fees. The code is on GitHub, and it’s what your automations are built on.

The services you buy from us run on a foundation you own, not ours. You’re not paying for a vendor’s AI. You’re paying for the team that deploys, integrates and operates AI on your terms.

How we work

Four phases, in the order your procurement team will recognise.

Start with a free review. Each phase ends with something you keep, and you can stop after any of them.

Architecture review

Free · 30 minutes

For the compliance lead, IT lead or senior stakeholder who needs to know whether this is real before committing to anything.

  • A working session with a senior engineer from the team that builds Open Assistant
  • A map of where AI is used in your organisation today and where the data flows
  • Which use cases your current policy blocks, and which become possible on self-hosted infrastructure
  • Plain-language answers to your DPO's and CISO's questions
OutputA one-page deployment plan: what becomes possible on your infrastructure, the shape of the cost, and what the next phase involves.
Book the 30-minute review →

Compliance proof

Paid · fixed scope, agreed upfront

For teams that have done the review and need a working pilot plus the documentation to get internal sign-off.

  • Open Assistant deployed on your infrastructure: your VPS, your on-prem hardware or your own EU cloud
  • The integrations you need configured (Outlook, calendar, CRM, internal tools)
  • Operational documentation written for an auditor, not an engineer
  • Technical inputs for your DPO’s privacy impact assessment
OutputA working pilot your team can use, plus the audit trail that got it approved by your compliance function.
Discuss a pilot →

Team rollout

Paid · days, not months

For teams with a working pilot that need it in real workflows, with day-to-day operation handed to an internal owner.

  • Workflow integration: we map the manual processes the assistant takes over (an AI scheduling assistant instead of calendar back-and-forth, for example) and build the automations alongside your team
  • Custom integrations if needed, shipped on the same software rather than as forks
  • Training for the internal admin who will run it after we leave
  • Handover documentation covering routine operation, upgrades and troubleshooting
OutputA self-hosted AI assistant your team uses daily, run by an internal admin you trust.
Plan a rollout →

Managed operation

Optional · monthly, cancel anytime

For organisations that do not want to own day-to-day operation but still want the assistant under their control.

  • The assistant runs on your infrastructure, or on EU hosting in your name if you prefer, monitored and patched by us
  • Updates shipped with a change-log your compliance team can review before they go live
  • A named point of contact for incidents, escalations and roadmap input
  • Stopping the service deletes nothing: the database, code and credentials stay yours
OutputSovereign AI in production, operated by the team that builds the software, on hardware you control.
Talk to the operations team →
What’s in the foundation

Specifics, not adjectives.

Source-available · BSL 1.1

A codebase your developers can read. Verify exactly what runs where.

View on GitHub →

Self-hosted by default

Runs on your VPS, your on-prem hardware or your own cloud. It does not need us to host it. See the guide to self-hosted infrastructure.

88+ tools

Outlook, Gmail, Google Calendar, Notion, Slack, CRMs, project tools and custom APIs.

Browse the docs →

EU data residency

Configure for the region you need. Nothing leaves your perimeter unless you decide it should.

No per-execution fees

Flat infrastructure cost. No vendor metering on every prompt.

Model-agnostic

Run open-weight models (Llama, Qwen, Mistral, DeepSeek) or bring your own API keys.

In practice

What it looks like when the constraint goes away.

Three scenarios from the teams we talk to most. Each links to the full write-up.

Regulated team wants AI for email triage, but its data can’t leave the EU

A compliance team has blocked every AI tool that routes corporate email to a US-hosted vendor, and has been triaging shared mailboxes by hand for over a year.

Open Assistant runs on a VPS in Frankfurt. The Outlook integration reads shared mailboxes, triages inbound mail, drafts replies in the team’s tone and surfaces only what needs a human. Nothing leaves the EU.

Read the full post →

A small legal or financial practice pays per seat for an inbox filter it can’t audit

A 12-person firm has been paying for SaneBox and cannot verify where its metadata goes. It has been told the EU AI Act will tighten requirements next year.

Open Assistant runs on a single VPS in Belgium: the same triage outcome, full data residency, an audit trail, and no per-seat fee that grows with the team.

Read the full post →

A dev team needs AI in its workflows, but security won’t approve a cloud SaaS

The team uses n8n today but wants AI-native automations rather than node pipelines. Its security team has rejected every cloud AI vendor.

Open Assistant replaces the workflow layer with an agent that executes on the team’s own infrastructure: 88+ tools, no visual pipeline to maintain, no data leaving the corporate perimeter.

Read the full post →
What makes this different

Honest about the alternatives.

Vendor-hosted tools. Most “AI for business” products are SaaS hosted in the US. You pay per seat, your data flows through their infrastructure, and your compliance team’s audit trail is whatever their policy document says it is. If your constraint is “data must not leave our perimeter”, these tools do not solve it.

Doing it yourself. You can absolutely self-host an AI assistant yourself, and many teams do. The engineering work is real: installation, integrations, security hardening, model selection, operations. Our services are for teams who want the self-hosted outcome without staffing an internal platform team.

Agencies. Pure-services AI agencies build automations on top of someone else’s AI vendor. You get the productivity gain, but you still don’t control the foundation. If the vendor changes pricing or terms, or disappears, your automations are exposed.

What we do. We deploy and operate Open Assistant, an open codebase on your infrastructure. The services you buy run on software you control.

FAQ

The questions that stop most buyers.

Where does our data actually go?
Where you decide. By default, Open Assistant runs on infrastructure you choose: your VPS, your on-prem server or your own EU cloud. We deploy it there and we do not route your data through our systems or log your prompts and responses. If you would rather have us operate it, it runs on EU infrastructure in your name.
Can our compliance team audit this?
Yes. The code is source-available under BSL 1.1 and can be read and verified by your developers. We provide deployment documentation written for an auditor, not an engineer, and every tool invocation is recorded in an audit log. If your compliance team needs a session with our engineering lead, that is included in the compliance proof phase.
What models does it run?
Open Assistant is model-agnostic. You can run open-weight models (Llama, Qwen, Mistral, DeepSeek, Gemma) entirely on your infrastructure with no external API calls, or bring your own API keys for commercial models. You decide.

Read the model documentation →

What happens if you disappear?
Nothing breaks. The software is on your hardware, the database is a file you can copy, and the code is available to you under BSL 1.1. Your internal admin owns day-to-day operation by the end of the team rollout. We are a services partner, not a dependency.
How is pricing structured?
The 30-minute architecture review is free. The compliance proof and the team rollout are fixed-scope engagements, scoped and priced before anything starts. Managed operation is an optional monthly fee that you can cancel at any time. The software has no licence fee and we do not meter prompts or executions; you pay your infrastructure and model provider for what you use.

Ready to see Open Assistant running on your stack?

The 30-minute architecture review is the fastest way to find out whether a self-hosted AI assistant solves the constraint your team has been routing around.

Book the architecture review → Read the private-AI-assistant guide first →