On-premise, in the cloud, or run by us
Every workflow we build can run on your own hardware, in a dedicated cloud environment under your keys, or as a standing agent we host and keep running. The same questions decide which; the assessment answers them for your situation.
Three places it can run
The workflow is the same. What changes is where the box sits, who holds the keys and who keeps it running.
You run it
On-premise
Your hardware, your network, optionally air-gapped.
- · Data that may not leave the network, by contract or policy
- · High, steady volume where owned hardware pays off
- · A team that already runs its own servers
You own it, we build it
Cloud
A dedicated environment in your cloud account, under your keys.
- · Regulated data that must stay in a named region, without buying hardware
- · Variable or growing volume
- · No infrastructure team, or one that would rather not own GPUs
We run it
Managed agents
A standing agent we host, monitor and keep running. You get the report.
- · A report, brief or digest you assemble by hand today, on a cadence
- · Nobody on your side who wants to own prompts, retries and connectors
- · Sources that change often enough that maintenance is the real cost
What decides where it runs
Four questions that come up on the first call, and what each actually requires.
Data residency
Contracts, privacy law or client policy say where personal and commercial data may live. On-premise and cloud answer with a location you control; a managed agent answers with an agreed data-handling rule set and inputs minimised before any model sees them.
Who will operate it
Someone keeps it running: your team with runbooks, your cloud provider's tooling, or us. An honest answer to this question rules out at least one option before anything else is weighed.
Vendor exposure
A workflow built on one provider's API inherits its pricing, outages and policy changes. Open-weight models on infrastructure you control remove that dependency for the steps that matter.
Cost at volume
At high, steady volume, owned hardware or a dedicated environment can cost less than per-run pricing. At low or bursty volume, a managed agent usually costs less than the people to run one. The assessment says which is true for you.
Side by side
The same criteria decide it every time. The assessment answers them for your situation.
| Criterion | On-premiseYou run it | CloudYou own it, we build it | Managed agentsWe run it |
|---|---|---|---|
| Data sensitivity | Data that may not leave the building or the network | Regulated data in your own cloud account, under your keys | Data we may process under an agreed rule set, minimised before any model sees it |
| Volume | High and steady; hardware is sized once | Variable or growing; the environment scales up and down | Any; runs are scheduled and metered |
| Team | Someone in-house owns the hardware and the network | No infrastructure team required; you or your cloud provider operate it | No one on your side; we operate it and send a monthly health note |
| Budget shape | Capital up front, lower run cost after | Monthly infrastructure spend, scaled to use | A fixed build plus a monthly fee |
| Latency | Local and deterministic | A network hop to your cloud region | Not a factor; outputs arrive on a schedule |
| Compliance | Your controls end to end, air-gapped if needed | Cloud provider certifications plus your controls | Our controls, documented; your approval before any write |
| What you see | Everything; the deployment is yours | Everything; the deployment is yours | The report and its sources; not the pipeline |
What stays constant
Whichever option runs it, four things do not change.
Models chosen by measurement
On-premise and cloud run open-weight models you can inspect and pin, served with Ollama or vLLM. A managed agent uses the model that measured best on your task, named in the scope.
Citations
Outputs cite the document or record they came from, whatever runs them.
Human approval
Nothing writes to a system of record without a person approving it.
Evaluation gates
A frozen evaluation set decides whether a change ships, in every option.
Canadian data residency
- Where your obligations require it, data stays in Canada: your own premises for on-premise, a Canadian cloud region for cloud. For a managed agent, the region and the retention rules are written into the data-handling rule set before the first run.
- We design for PIPA and PIPEDA obligations and say where the design meets them; certification and legal opinion stay with your counsel.
- The residency map is a deliverable of the assessment, not an assumption.
How the engagement runs
- 1
Assessment
Data sensitivity, volume, team, budget, latency, compliance. A written recommendation: on-premise, cloud, a managed agent, or none of the three.
- 2
Reference architecture or scope
For on-premise and cloud: components, data path, keys, backups, updates, cost shape. For a managed agent: what it reads, what it produces, how often, who receives it. Reviewed with your team before anything is built.
- 3
Build
Containers, a vector store, a model server, the workflow. Run beside the current process until the outputs match.
- 4
Handover or run
On-premise and cloud: documentation, runbooks and the evaluation set, with an optional run agreement. Managed: we keep it running and send a monthly health note.
What this is not
The TurnkeyGTM platform runs as a cloud service we operate. We do not offer an on-premise or private-hosted edition of the platform today. On-premise and cloud deployments are designed and built on open-weight models and self-hostable components — containerised services, a vector store such as Qdrant or Postgres, and a model server such as Ollama or vLLM — and handed over with documentation. Managed agents run on infrastructure we operate, under a data-handling rule set agreed before the first run. Capacity, latency and cost figures are given per engagement, never on this page.
Questions we get asked
The capabilities behind a deployment
Custom hosted and on-premise AI
AI deployments designed for where your data has to live — a dedicated hosted environment or your own hardware — on open-weight models and self-hostable components.
See the deliverablesAI workflow automation
The workflow your team repeats every week, automated with AI where it helps and plain software where it does not, run beside your current process until you trust it.
See the deliverablesPII protection and data handling
Redaction, minimisation and handling rules for personal data in transcripts, tickets and documents before any model sees them, with residency where your obligations require it.
See the deliverablesAI policy and governance advisory
Acceptable-use, model-risk and data-handling policy for a team facing client, insurer or regulator questions.
See the deliverablesTell us where your data has to live.
The assessment is short and fixed price. It ends with a written recommendation: on-premise, cloud, a managed agent, or none of the three.