Context
Context management
The context layer is being developed to manage the information required for a task across local interactions. Its responsibilities include maintaining relevant working context while respecting device memory limits.
Technology
Engineering a complete
local AI system.
GreenTrust develops an on-device inference architecture combining context management, model selection and resource-aware execution. The work is currently focused on achieving reliable operation under the practical limits of personal computing hardware.
Models
Specialised models and model selection
Our model development strategy combines in-house work with the adaptation of existing models where appropriate. We investigate how specialised components can be selected according to task requirements and available resources.
Compute
Resource-constrained inference
Model residency, memory use, latency and energy requirements are engineering constraints. We evaluate the trade-offs between resource consumption and task quality to inform execution decisions.
Locality
Fully on-device execution
The planned system runs model inference locally; it does not use hosted AI inference endpoints. Working context and task execution are designed to remain on the user’s device.
Trust
Verification and reliability
We aim to make system behaviour measurable through reproducible experiments, explicit operating constraints and documented limitations. Technical experiments inform reliability and verification priorities.
Principle
The execution boundary
is the device.
Implementation choices are being tested and refined, while the requirement that inference remain fully local is fundamental to GreenTrust.