Anvia
An open-source TypeScript agent runtime for production AI systems, with typed tools, structured context, provider-neutral models, retrieval, workflows, streaming, and observable execution.
Highlights
- Created an open-source provider-neutral runtime boundary for AI applications.
- Designed typed primitives for tools, context, retrieval, workflows, and streaming.
- Kept application infrastructure owned by the product team.
Context
Anvia is my attempt to define a clearer runtime layer for AI products. Many agent systems begin as a small wrapper around model calls, then become hard to inspect once tools, retrieval, memory, workflows, and permissions enter the system.
I wanted an explicit TypeScript runtime that lets teams model those pieces directly instead of hiding them behind loosely shaped callbacks.
What I focused on
The main design decision was to keep Anvia focused on the AI runtime boundary. It can coordinate models, typed tools, structured context, retrieval, workflow execution, and observability without becoming the source of truth for auth, billing, users, or product data.
That boundary makes the runtime useful in production systems where teams need control, auditability, and incremental adoption.
Outcome
Anvia gives product engineers a stronger foundation for building AI systems that need to be tested, observed, and evolved over time.