For the Developer
Sovereign intelligence for the Developer
Build and deploy AI without giving up control of your stack.
Developers feel lock-in first, because it arrives as a proprietary SDK, a hidden rate limit or a prompt format that only works in one place. You want to ship quickly without making choices you cannot reverse, and you want to be able to explain to your own security and finance colleagues what the stack does and what it costs. This page lists what to look for and the questions that reveal how portable a platform really is.
What a Developer worries about
Proprietary SDKs and formats
If your agents, prompts and workflows only run inside one vendor's framework, every future change is a rewrite. Open interfaces keep options open.
Hidden limits and unpredictable cost
Rate limits, context-window pricing and retries can break a prototype at scale. You need visibility into cost per call and a way to route to cheaper models. See model routing and cost.
Debugging agents
When an agent loops or calls the wrong tool, you need a trace of each step with inputs and outputs. Without it, debugging is guesswork.
Security as an afterthought
Tokens in environment files, tools with broad scopes and unreviewed MCP servers are common. Secure defaults save you from retrofitting them.
Local and offline development
Teams want to develop against local models and test without sending data anywhere. A platform that only works in the vendor's cloud slows iteration and raises risk.
What a Sovereign Intelligence Platform gives you
One API across many models
Connect is an entry point for a single API to 50+ model providers with routing, failover and cost tracking, so switching models is a configuration change.
Build agents with or without code
Studio builds agents, chatflows and automations without code, and BuildX orchestrates multiple agents that develop software from natural-language descriptions.
Policy and audit that apply to your code
Nexus is designed to capture agent actions and enforce policy in flight, so governance applies to agents regardless of how they were built.
Local-first options
Cortex runs on the laptop by default, with on-device models, which gives a way to work with sensitive content without it leaving the machine.
A start that does not close doors
You can sign up, build and then grow into private or dedicated deployment as needs change. See the guide for the sequence and pricing for plans.
Capabilities are described as what the platform is designed to let you do. For what is true today and what is not claimed, see the trust centre.
Questions to ask any vendor
Use this as a checklist in any evaluation, ours included. Each question comes with what a good answer looks like.
01Is the API compatible with common client libraries, and can I swap providers by changing configuration?
A good answer: A documented, stable API surface compatible with widely used client conventions, with provider selection in config rather than code.
02Which open protocols do you support for tools and agents?
A good answer: Support for the Model Context Protocol and similar open standards, with version numbers. MCP is now governed under the Linux Foundation (opens in a new tab).
03Can I run the core runtime myself, and under what licence?
A good answer: Self-hosting is available for defined components, with licence terms stated and clarity about what remains vendor-hosted.
04How do I develop and test locally, including against local models?
A good answer: A local mode or emulator, support for local model runtimes and fixtures for tests without live data.
05What tracing do I get per agent step, tool call and model call?
A good answer: Spans with inputs, outputs, latency, token counts and cost, exportable in an open format. The OpenTelemetry GenAI conventions are in Development status, so confirm which version is used.
06How do evaluations fit into CI?
A good answer: A CLI or API to run evaluation suites on pull requests and fail builds on regressions.
07How are secrets and tool permissions handled for agents I build?
A good answer: A secrets manager, scoped per-agent tokens and a policy layer so a tool call can be denied without changing agent code.
08What are the rate limits, retries and timeouts, and how do I see them?
A good answer: Documented limits, headers or metrics that expose them, and configurable retry and fallback behaviour.
09How do I export my agents, prompts, workflows and logs?
A good answer: Export via API or CLI in documented formats, with no dependency on a support request.
10Do you train on my prompts, code or outputs?
A good answer: A contractual yes or no covering prompts, files, outputs and feedback.
11How do I put guardrails and approvals in front of an agent without forking its code?
A good answer: Policy is defined outside the agent and evaluated on each model or tool call, so an allow, deny, filter or approval rule can be changed through configuration, versioned in source control and tested like any other change, rather than being baked into prompts.
12What happens to my app when a model is deprecated?
A good answer: Advance notice, version pinning, a migration guide and tested replacements through the same API.
Recommended reading
- Guide: select and evaluate models
- Guide: build governed agents
- Model sovereignty
- MCP and agent interoperability
- Sovereign AI stack, layer by layer
Not sure where to start? Take the readiness assessment or read the build guide.
Frequently asked questions
How do I start building?
Use Build with AI to sign up, pick one entry point such as a model API or an agent, and add governance and infrastructure as the use case grows.
Can I use open-weight models?
The platform is designed to route across many models, including ones you host. See model sovereignty for how to choose.
Does governance slow development?
It should not. Policy sits outside agent code, so you change what an agent may do without redeploying it, and you start restrictive and relax with evidence.
Build with control: for the Developer
Start with one entry point. Add intelligence, agents, workflows and infrastructure as you prove value. Or read the step-by-step build guide and take the readiness assessment.