Timaintechnical documentationTechnical documentation
How the platform works underneath, for engineers, platform teams and security reviewers. These pages describe what is built, name what is not, and avoid claims that have not been demonstrated.
Where to go
Delivery path
The seven steps from intent to a verified deployment, and what the platform never does.
Agents
Personas as contracts: context, tools, budgets, model policy and pushback.
Security
The trusted kernel, authorization, tenant isolation, sandboxing and signed evidence.
Status and limits
What runs today, what is in first runs, what is planned, and the known limits.
Agents that are bounded, not unleashed
The platform borrows the structure of a good software organisation. Work is directed by an architect, carried out by specialist personas, and completed only by the platform itself.
The Architect principle
Every change starts as a design decision. The architect turns approved requirements into a project revision and a plan of typed work, so no agent improvises the shape of the system.
A team of personas
Product, architect, implementation and verification roles each have their own contract, context, tools and budget. A persona can only do what its role is for, and a verifier never trusts the author.
Controllers complete, agents propose
An agent asks for validation; the platform checks the evidence and decides. No agent can mark its own work done, certify a release or reach production on its own say-so.
Pushback is a setting
Tenants choose how hard the product and architect roles challenge scope creep, gold plating and weak assumptions, and every concern they raise is recorded and acknowledged by a person.
Best-practice SDLC, built in
Evidence, not claims
Builds, tests, contract checks, images and deployments each produce signed receipts. A change is delivered when the required deployment is verified, never when code merely exists.
Reproducible everywhere
Generated code with provenance, reproducible images pushed by digest, and environments created from scratch. The same artifact that passed certification is the one that ships.
Sandboxed by default
Builds and tests run in isolated containers with no network and no host access. Secrets never enter prompts, artifacts or logs.
Upgrades without fear
Every upgrade is just another change through the same loop, so the organisation's confidence in the first delivery carries to the hundredth.
Usage-based and transparent
Usage is metered per tenant and project. No price list is published.
Every meter, per tenant
AI spend, build time, environment time and storage are metered for each tenant and project, and every figure traces back to the work that caused it.
Budgets stop spend before it happens
Set a budget and the platform holds work back before it spends past it, then resumes when the budget is raised. Nothing is billed for work that was refused.
Environments
Disposable. Created from scratch for each change and torn down when the change is proven.
Fresh for each candidate release, seeded with a versioned dataset, certified by scenarios that actually ran.
Receives only certified releases, under current policy, with a human approval bound to the exact release.