Memory

Ports of the .NET memory samples. Step numbers follow upstream.

Example	Demonstrates
Step 01: Chat history memory	Remembering a user's preferences across two independent sessions
Step 05: Bounded chat history	Keeping four recent messages and recalling older messages from vector-store overflow

Both use memory.HistoryProvider and an in-memory vector store. Memories survive new sessions, but not process restarts. The fixed UID1 scope is for demonstration; applications should use the authenticated user's identifier.

Run

Sign in with a credential supported by DefaultAzureCredential, such as az login, and configure deployed Foundry models:

$env:FOUNDRY_PROJECT_ENDPOINT = "https://<resource>.services.ai.azure.com/api/projects/<project>"
$env:FOUNDRY_MODEL = "gpt-5.4-mini"
$env:FOUNDRY_EMBEDDING_MODEL = "text-embedding-3-large"

go run .\examples\02-agents\memory\step01_chat_history_memory
go run .\examples\02-agents\memory\step05_bounded_chat_history

These samples use 3072-dimensional embeddings, matching the shared demo default, text-embedding-3-large. If your embedding deployment produces a different vector size, update the dimension argument to memory.NewHistoryProvider in the sample. Runs invoke live services and may incur charges.

Step 05 preserves the first system message outside the four-message window. Function-call and function-result messages are excluded from both retained history and archived memory. Retrieved memories are prepended to recent history and are not stored again.

Local validation

go build .\examples\02-agents\memory\...
go vet .\examples\02-agents\memory\...