Intelligent systems

Agents, model runtimes and consented learning, built to do real work inside the platform rather than to demonstrate in isolation.

System map

AI systems

Developer Preview

Purpose

Agents that plan, act, validate and report: connected to real project work, not isolated demos.

Architecture

Agent framework + AI runtime, running compatible models across cloud, desktop, edge and RISC-V surfaces.

Each has a defined job and a different stage. Two are usable in developer preview; two are earlier.

AI Agents, Developer Preview: Plan, act, validate and report against a real project, with the engine in the loop.
Learning Loop, Developer Preview: Capture outcomes locally, redact secrets, and improve future work only with consent.
AI Runtime, In Development: Run compatible models across cloud, desktop, edge and open architectures through one interface.
Multimodal models, Research: General-purpose models spanning language, image, audio, video and structured data.

The commitments that shape the design, stated as design properties rather than policy language.

Local first: Traces are captured on your machine. Nothing leaves it by default.
Consent gated: Sharing is opt-in per decision, not a single setting buried at install time.
Secrets redacted: Credentials and keys are stripped before anything is written to a shareable trace.
Provider keys stay yours: Model access uses keys held in a local vault; the project uses your keys, not a shared account.