Provider-agnostic agents, tool workflows, and structured extraction for TypeScript applications.

Anvia is a TypeScript runtime for building provider-agnostic agents, tool workflows, and structured extraction inside your application code.

It is designed for teams that want more structure than raw model calls, but less framework weight than a full orchestration stack. Provider clients create models. Builders configure behavior. Your application still owns data, permissions, persistence, and side effects.

Anvia gives you a compact set of primitives for production-adjacent AI workflows:

• Agents, tools, extractors, pipelines, streaming, and typed output schemas.
• Provider-backed completion and embedding models across OpenAI, Anthropic, Gemini, Mistral, and compatible APIs.
• RAG primitives with in-memory search, local embeddings, ChromaDB, Qdrant, and pgvector adapters.
• Runtime integrations for MCP, local skills, multimodal attachments, observability, and Anvia Studio.

Choose Anvia when you are building AI features inside a TypeScript product and want provider-agnostic agents, extractors, tools, pipelines, and Studio workflows without giving up control of your application's data, permissions, persistence, and side effects.

For deeper positioning, see Comparison and Design Philosophy.

import { AgentBuilder, ExtractorBuilder, PipelineBuilder } from "@anvia/core";
import { OpenAIClient } from "@anvia/openai";

const client = new OpenAIClient({
  baseUrl: "https://openrouter.ai/api/v1",
  apiKey,
});
const model = client.completionModel("qwen/qwen3.6-35b-a3b");

const agent = new AgentBuilder("support", model)
  .instructions("Answer support questions clearly.")
  .build();

const response = await agent.prompt("How do I reset my password?").send();

const extractor = new ExtractorBuilder(model, ticketSchema).build();
const ticket = await extractor.extract(response.output);

const workflow = new PipelineBuilder<string>()
  .step((input) => `Summarize this support ticket:\n\n${input}`)
  .prompt(agent)
  .extract(extractor)
  .build();

const normalizedTicket = await workflow.run(
  "Acme Co. reports checkout failures. Priority is high.",
);

Install dependencies:

pnpm install

Create a local .env file for cookbook runs:

OPENROUTER_API_KEY=...
OPENAI_API_KEY=...
GEMINI_API_KEY=...
MISTRAL_API_KEY=...

Run the first basic text call:

pnpm cookbook:basics:01

Run Studio locally:

pnpm cookbook:studio:01

Start the cookbook ChromaDB service before running the Chroma-backed RAG examples:

docker compose -f examples/cookbook/compose.cookbook.yml up -d
pnpm cookbook:retrieval:05
pnpm cookbook:retrieval:06
pnpm cookbook:retrieval:07
pnpm cookbook:retrieval:08

The cookbook is a product learning path. It starts with a plain text call, then adds history, streaming, tools, extraction, providers and multimodal APIs, pipelines, retrieval, multi-agent workflows, evals, Studio, and integrations one concept at a time.

Run the default example for a level:

pnpm cookbook:basics
pnpm cookbook:tools
pnpm cookbook:structured-output
pnpm cookbook:providers
pnpm cookbook:pipelines
pnpm cookbook:retrieval
pnpm cookbook:multi-agent
pnpm cookbook:evals
pnpm cookbook:studio
pnpm cookbook:integrations

Numbered scripts are available for each level when you want to step through the path in order, for example pnpm cookbook:basics:01. Existing basic, intermediate, pipeline, rag, and multimodal cookbook scripts remain as compatibility aliases.

Install dependencies before running workspace tasks:

pnpm install

Common commands:

pnpm typecheck
pnpm test
pnpm build
pnpm check

Package-scoped commands:

pnpm --filter @anvia/core typecheck
pnpm --filter @anvia/core test
pnpm --filter @anvia/core build

pnpm --filter @anvia/studio typecheck
pnpm --filter @anvia/studio test
pnpm --filter @anvia/studio build

pnpm --filter cookbook typecheck

.
├── packages/
│   ├── core/                         # @anvia/core
│   ├── providers/
│   │   ├── openai/                   # @anvia/openai
│   │   ├── anthropic/                # @anvia/anthropic
│   │   ├── gemini/                   # @anvia/gemini
│   │   └── mistral/                  # @anvia/mistral
│   ├── vector-stores/
│   │   ├── chroma/                   # @anvia/chroma
│   │   ├── qdrant/                   # @anvia/qdrant
│   │   └── pgvector/                 # @anvia/pgvector
│   ├── observability/
│   │   ├── langfuse/                 # @anvia/langfuse
│   │   └── otel/                     # @anvia/otel
│   ├── embeddings/
│   │   ├── transformers/             # @anvia/transformers
│   │   └── fastembed/                # @anvia/fastembed
│   └── tools/
│       └── studio/                   # @anvia/studio
├── apps/
│   └── docs/         # documentation app
├── examples/
│   ├── cli-agent/    # runnable CLI agent example
│   └── cookbook/     # runnable examples
├── biome.json
├── pnpm-workspace.yaml
└── tsconfig.base.json

Keep changes small and covered by the relevant package tests. For API changes, add or update cookbook coverage so the behavior is easy to verify from the command line.

Before opening a change, run:

pnpm typecheck
pnpm test
pnpm build
pnpm check

MIT.