Examples
Capability Map
Map Anvia features to the examples that show them in application flows.
Use this map when you already know which Anvia feature you want to use and need to find the example that places it in an application flow.
The concrete RAG stack used in examples is Mistral OCR, OpenAI embeddings, and Chroma. Treat that as the reference path; provider and store pages show the swappable alternatives.
Core Runtime
| Capability |
Example flow |
What the example should prove |
| Agents |
Agent App Flow, Support Agent |
A product request can run through auth, tools, model calls, traces, and persistence. |
| Runtime composition |
Agent Runtime Composition |
Model, instructions, tools, context, memory, observers, and output contracts are assembled at the right boundary. |
| Request runners |
Agent App Flow, Pipeline Worker |
Routes, jobs, and tests call the same workflow function. |
| Memory and sessions |
Memory and Events, Prisma Agent Memory, Drizzle Agent Memory, Raw SQL Agent Memory |
Conversation messages are loaded and appended through scoped application storage adapters. |
| Streaming events |
Fullstack Streaming, Streaming Events |
Browser clients consume completion or agent streams, then product UIs treat text, tools, and final events as workflow state. |
| Event store |
Memory and Events, Prisma Agent Event Store, Drizzle Agent Event Store, Raw SQL Agent Event Store |
Runtime stream events are persisted by run id for replay, debugging, and audit. |
| Capability |
Example flow |
What the example should prove |
| Typed tools |
Permissioned Tools |
Tool schemas describe the model-facing contract while services enforce product permissions. |
| Tool authorization |
Permissioned Tools, Tool Validation |
User and tenant scope come from the request, not model arguments. |
| Tool approvals |
Guarded Side Effects, Human Input |
Sensitive work pauses or routes to a human decision before execution. |
| Tool output safety |
Tool Validation, Runtime State and Persistence |
Tools return narrow, safe result shapes and keep private records in application storage. |
| Dynamic tool catalogs |
Dynamic Tool Catalogs |
Large tool sets are searched and filtered before the model sees them. |
| MCP tools |
MCP Agent |
External server tools are filtered and wrapped before joining app-owned tools. |
Knowledge And Retrieval
Structured Workflows
Production
| Capability |
Example flow |
What the example should prove |
| Tests |
Testing Harness |
Deterministic boundaries are testable without relying on live model behavior. |
| Traces and observers |
Observability Loop, Runtime State and Persistence |
A run can be debugged from request id to model events and tool calls. |
| Evals |
Eval Loop |
Completions, parsed completions, streams, agents, and product runners become repeatable regression targets. |
| Provider switching |
Provider Switching |
Model selection changes at the runtime boundary, not throughout product code. |
| Launch review |
Production Readiness |
The workflow has permission, privacy, retry, audit, and operational checks. |