Aiva model family
Intelligence for every private task.
Six purpose-built Aiva models work as one assistant, combining rapid everyday helpers with deep synthesis, research, and agentic orchestration.
Aiva Models
Secure
Purpose-built
Family of models
Private by design
Models run in the boundary you choose.
Explicit by selection
Choose a specific model you want to use.
Built as a family
Each model with distinct level of brilliance.
The family
Six models. One intelligent assistant.
Aiva provides models as compatible choices inside one experience, so you can lean on speed for everyday tasks and step up to reasoning or orchestration when the work calls for it.
Capability architecture
Each model protects the experience from excess.
Aiva can stay quick when the work is small, while making deeper synthesis available on demand.
Local response layer
Spark and Atom handle low-latency interactions, quick transformations, and immediate workspace actions.
Grounded workspace reasoning
Pulse and Logos provide deep context window synthesis, structured document analysis, and rigorous argument construction.
Deep synthesis and orchestration
Sage and Nexus deliver high-context research, policy-heavy evaluation, and autonomous multi-agent task execution.
One assistant, many modes
Seamlessly shift between model profiles without changing tools or exposing workspace boundaries.
Model routing
The right amount of intelligence for the job.
Aiva's model family is designed around judgment: matching model capacity to task scale so you get rapid responses for simple requests and deliberate synthesis when complex decisions depend on it.
Start close
Spark, Atom
For quick private work, low-friction drafts, and everyday desktop tasks.
Deepen when useful
Pulse, Logos
For documents, structured reasoning, synthesis, and workspace analysis.
Escalate with intent
Sage, Nexus
For expert work, policy review, deep research, and multi-step orchestration.
Model selection never changes Aiva's central promise: your workspace stays under your control, with local inference by default.
Aiva experience