Frequently asked questions
Understand what Anvia provides, where it fits in a TypeScript application, and how it differs from other AI development tools.
Start with your question
| Question | Answer |
|---|---|
| What is Anvia? | A composable TypeScript runtime for provider-neutral AI behavior inside application code. Learn the boundary. |
| Why use it? | To share explicit contracts across models, agents, tools, memory, retrieval, pipelines, streaming, and observability without giving a framework ownership of the surrounding application. Why Anvia. |
| How does it compare with other TypeScript tools? | The best choice depends on whether the product needs a model/UI toolkit, an integrated agent framework, an autonomous harness, or an application-owned runtime. Compare the stacks. |
| Do I need Studio or Lens? | No. Both are optional products around the SDK: Studio is a local development console, while Lens is an observability and evaluation workspace. Understand the products. |
| Is it production-ready? | The runtime provides production-oriented controls, but the application still owns deployment, authentication, authorization, data policy, and operational reliability. Read the production answer. |
| Does it work with Bun? | Node.js is the currently tested and supported runtime. Bun may work for some packages, but official support will follow dedicated compatibility testing. Read the Bun answer. |
Explore by topic
Understanding Anvia
Start with what Anvia is, how it is structured, and what the application still owns. The boundary is as important as the feature list.
Choosing a stack
Read the capability overview, then compare Anvia with Vercel AI SDK, Mastra, VoltAgent, Flue, or direct provider SDKs.
Runtime capabilities
Decide between completions and agents, memory and knowledge, or pipelines and agents. Separate pages cover tools, multi-agent systems, structured output, streaming, and sandbox execution.
Studio and Lens
Learn what Studio and Lens are designed to do—and why Studio is not a production operations console.
Production architecture
Review security ownership, deployment, Bun support, persistence, retries and cancellation, and testing agent behavior.
Quick answers
Is Anvia tied to OpenAI?
No. Core depends on provider-neutral model contracts. Provider packages adapt OpenAI, Anthropic, Gemini, Mistral, Grok, and supported compatible endpoints. Model capabilities still vary by provider, account, region, endpoint, and model ID.
Do I need an agent for every model call?
No. Use a direct completion for one model request. Use an agent when reusable behavior needs tools, context, memory, hooks, or multiple bounded turns.
Does a tool schema provide authorization?
No. A schema validates the arguments a model produced. Application code must authenticate the caller and authorize the requested operation.
Does a memory scope authorize a session?
No. Scope determines which conversation is loaded. The application must still verify that the caller can access that user, tenant, and session.
Where should provider keys live?
On the server or trusted worker. Browser clients should call an authenticated application route that runs the model or agent and returns an appropriate stream.
How do I inspect an agent locally?
Register it with @anvia/studio, start the local runtime, and inspect its tools, context, sessions, traces, and configured capabilities in the browser.