Layer 03
Intelligence & Models
Turn data and knowledge into useful intelligence.
No single model fits every task, and no organisation should be tied to one. This layer lets you choose, route, evaluate and tailor models, from foundation models to open and private ones, and apply the same governance whichever you pick.
What the models layer includes
- Foundation models
- Open models
- Private models
- Tailored models
- Fine-tuning
- Routing
- Inference
- Evaluation
- Selection
- Optimisation
A model estate you can manage
Foundation models
Frontier models from leading providers, reached through one API.
Open models
Open-weight models that you can host on infrastructure you control.
Private models
Models deployed only for your organisation, with their own access and retention rules.
Tailored models
Models adapted to your domain through fine-tuning and optimisation.
Selection, routing and evaluation
Choosing a model is a continuing decision, not a one-off purchase.
- 01
Select
Approve the models each use case may use, by risk level and data classification.
- 02
Route
Send each request to the model that fits cost, latency, availability and policy.
- 03
Evaluate
Compare models on your own tasks before and after they reach production.
- 04
Optimise
Adjust routing, caching and tailoring as models and needs change.
What Connect does today
Connect is Swfte’s universal model gateway. It gives access to 50+ model providers through one API, with smart routing, automatic failover, cost tracking and observability. It is the practical entry point into this layer, and the place where approved-model policy can be applied to every request.
Connect is also the routing layer that agents in the next layer call. It can be used as a managed service, or deployed in your own cloud, VPC or data centre (designed for, on request). See Connect self-deploy, Connect providers, EU-first routing and Connect security for the detail.
Model sovereignty
Model sovereignty means controlling selection, deployment, customisation and lifecycle. Because routing sits behind one API, changing a model does not mean rebuilding the agents and workflows that depend on it.
Where this sits in the closed intelligence loop
Control, then intelligence, then agency, then execution, then outcomes. Outcomes and their evidence flow back into data and context, so the loop closes.
- 01 · Layer 01ControlDecide where and how AI runs.
- 02 · Layers 02 and 03IntelligenceTurn data and knowledge into useful intelligence.(this page)
- 03 · Layer 04AgencyLet AI act within defined authority.
- 04 · Layer 05ExecutionEmbed that action in how the organisation operates.
- 05 · Layer 06OutcomesMeasure business results, and feed evidence back into data and context.
With data and context, this layer forms the Intelligence stage: turning what your organisation knows into AI that is useful.
Where Swfte products sit in this layer
Product names are kept. Each is an entry point into the platform.
Connect
One API for 50+ model providers, with smart routing, failover and cost tracking.
Connect: providers and BYOK
The providers and open-weight models Connect routes to, with your own keys if you prefer.
Connect: self-deploy
Run the model gateway in your own cloud, VPC or data centre, designed for on request.
GPU
Reference for the GPUs used to serve and train large language models.
What this means for your team
- CTO
- Deploy models, agents and workflows on infrastructure you control.
- Chief AI Officer
- Move from AI experiments to governed production AI.
- Developer
- Build and deploy AI without giving up control of your stack.
Frequently asked questions
What is LLM inference?
Inference is the work of running a trained model to produce an answer. Where it runs, on which hardware and under which controls determines cost, latency and how much of your data leaves your environment.
Can I host models privately?
Private and open models are part of the platform’s design, deployable on infrastructure you control. Talk to our team about the models and hardware that fit your workload.
Which models can I use?
Connect reaches 50+ model providers through one API. Which models a given team may use is set by policy, so approved models differ by use case and risk level.
How do I avoid lock-in to one model provider?
By routing through a gateway. Applications call one API, and the model behind it can change as quality, cost or policy change.
Does Swfte fine-tune models?
Tailoring models to a domain is part of this layer. The right approach, from retrieval to fine-tuning, depends on your data and requirements, and we scope it with you.
Across every layer
These ideas apply to every layer of the platform.
Controlled autonomy
Five levels, from AI that recommends to AI that adapts within limits.
Trust Profile
The record that describes what each AI system is and may do.
AI sovereignty
Seven kinds of control over your AI estate.
AI governance
Governance that runs inside AI, not beside it.
Swfte Intelligence Platform
Analyse and visualise your data, usage, agents and outcomes, then build from what you see.
Company brain
One governed place for everything your organisation knows, for every product to read from.
Custom models
Models adapted on your own data, evaluated before release and deployed under your control.
How Swfte builds with Cortex
An honest first-party account of how we build, review and approve work, labelled by status.
Governed agents
Identity, permissions, policy, approvals and audit for agents, with ten templates.
Compliance-approved workflows
Workflows with approval gates and evidence built in, with twelve templates.
Build on Intelligence & Models with Swfte
Start with one entry point. Add intelligence, agents, workflows and infrastructure as you prove value.