Engineering lane
Improve and repair Sentinel’s own code inside a separate engineering path.
IntentClarifyCandidateAccept
Sentrium / Sentinel
A program that
A program that
builds on intent.
Sentinel creates apps and projects from a described outcome. It is being designed to improve and repair itself while keeping context, control and evidence outside the AI worker.
Current status: MVP in late-stage bug testing. Built on five earlier prototypes, with Temporal already incorporated.
Current MVP
One builder. Two separate lanes.
Sentinel separates changes to its own code from the apps and projects it creates. The user describes the outcome; ambiguity is clarified; a candidate is built and tested before acceptance.
Apps / projects lane
Build independent apps and projects separately from Sentinel’s source code.
IntentClarifyCandidateAccept
In progress: direct activation into live code is not yet complete.
AI worker architecture
AI is the tool. Sentinel is the product.
Sentinel manages context and control. Models perform bounded tasks. The goal is to make model choice replaceable without giving the model continuing authority over the system.
Interchangeable models
Compatible models can be used through the AI Gateway rather than tying Sentinel to a single provider.
Fresh context per task
Each worker receives only the context needed for one task. The next task uses a new worker with fresh context.
Strict guardrails
Every worker receives a bounded task with defined permissions, action limits and approval gates.
Design principle: AI workers do not hold authority. Models can change; context, control and evidence stay with Sentinel.
End-to-end Recorder
Build it. Then interrogate it.
The Recorder links user intent, build activity and results to each app version. The Build Inspector is intended to answer questions from the selected version’s evidence and acknowledge anything that was not recorded.
Full build-and-inspect workflow remains in late-stage testing.
01 / INTENTIntent + questions
02 / WORKDecisions + changes
03 / PROOFTests + recovery
04 / RESULTVersion + outcome
BUILD INSPECTOR
“What changed? Why was this chosen? Which tests passed?”Answers are tied to recorded evidence for the selected version.Scaling infrastructure
Scaling starts with the current MVP.
Temporal is already incorporated. The next validation step is to stabilise the MVP, obtain professional advice, then host it for controlled multi-user testing and measure capacity.
Stabilise the MVP
Complete debugging of building, refinement and self-repair.
Professional advice
Validate security and expand the existing infrastructure for hosted, multi-user operation.
Host and test
Add online compute and hosting, then measure capacity with trusted testers.
Today: one laptop, with testing limited to one person at a time. Hosted deployment and concurrent-user capacity still need testing.
Reusable skills vision
Record the work. Reuse the skill.
The Recorder and usage statistics are intended to become the foundation for a linked, searchable skills library: use an existing skill, extend a relevant skill, or build what is missing.
User intentA requested outcome enters Sentinel.
→
Search + matchFind skills and dependencies relevant to the request.
Use an existing skill
Extend a relevant skill
Build what is missing
User experience vision
One interface. Many capabilities.
The long-term goal is for people to ask for outcomes rather than navigate separate everyday apps. Specialised tools may remain, but routine capability can sit behind one Sentinel interface.
Routine requests
Appointments, reminders and recurring personal tasks.
Useful outputs
Plans, diagrams, materials lists, spreadsheets, PDFs and purpose-built tools.
Use the screen you prefer
Apple, Android and PC, with custom screens and watch notifications as the platform develops.
On the device
- Keep private personal information local where appropriate.
- Retain essential processing and offline capability.
In the cloud
- Run compute-intensive work remotely.
- Store selected records encrypted and linked to an ID rather than a name.
These are design goals, not completed or independently validated security claims. Professional advice will define the security model and what must remain local.
Long-term integration
Own more of the delivery chain.
Start with the platform. As adoption grows, explore deeper ownership of the infrastructure that delivers the Sentinel experience.
01Platform + hosting
02Compute + data centres
03AI models
Camera + battery
Prioritise capture, battery life and everyday use.
Connectivity
Explore Wi-Fi, mobile and satellite fallback.
Right-sized local compute
Keep only what privacy and offline use require on the device.
Long-term ambition: purpose-built Sentinel devices delivering a better everyday experience at around one-third of a premium phone’s price. The price target is not yet validated.