Per-customer continual fine-tuning
Incrementally update customer-specific adapters from approved outcomes without forgetting prior capability — with an automated regression harness that verifies every improvement before it ships.
Research
How living models work.
We are a research lab first. We invent new model technology and prove it in real industries — five capabilities that compose into a living model.
The inventions
Five capabilities that, composed, produce a living model.
- 01
Reasoning
Models that plan over long horizons, work through multi-step problems, and check their own reasoning before they act.
- 02
Agents
Autonomous, multi-step action that drives the underlying software end-to-end, with human approval at the points that matter. The agent owns the workflow; people manage agents.
- 03
Continual learning
Methods to keep a model improving in deployment: it updates from real outcomes and holds on to prior capability, with an evaluation discipline that verifies every gain before it ships. That discipline is as much the work as the training.
- 04
Recurrent language models (RLM)
Persistent recurrent state for unbounded memory and genuinely long-horizon tasks. It carries the full history of a process in state, ready to act on any of it.
- 05
Streaming & live models
Always-on models that read a continuous data stream and act on it in real time. The unit of work is a live environment, watched and acted on without pause.
Composed, these are living models: persistent memory, continuous learning, real-time streaming, autonomous action — one object that runs 24/7 and gets better the longer it runs.
How the learning compounds
Applied research tracks that turn a customer’s approved outcomes into model improvements at production scale.
Autonomous skill extraction
Agents synthesize reusable procedures from completed multi-step work, verified before promotion. Procedural patterns generalize across customers; customer data never does.
Reward modelling & alignment
The approval workflow — approve, reject, edit — is the training signal. We align to each customer’s house style and risk appetite, and surface low-confidence work to a human rather than acting alone.
Open by default, where it helps
We run a dual track. Generic frameworks are open-sourced — for adoption and for credibility with the ML community. The enterprise agent, orchestration, and proprietary skills are gated behind commercial licensing.
Expect a measured cadence of technical writing and selective open-source contributions: enough to signal capability and attract people who want to work on hard problems, without giving away the compounding advantage.
Talk to the lab.
Design partnerships open in the coming months. Until then we’re glad to talk — whether you’re a researcher, a future partner, or just curious about living models.
Get in touch