Samples and notebooks
Clone the repository and follow Run locally for SDK, runtime, and environment setup. Run commands below from the repository root. Use synthetic development data for live examples.
Persistent Agent Memory demo
Teach a preference in Session A, recreate the agent over that same durable session, then recall the promoted preference in a new Session B for the same owner. Requires an Agent Memory store and a model credential.
./scripts/run-agent-memory-demo.sh teach
./scripts/run-agent-memory-demo.sh recall
# Or run both stages with bounded polling for memory promotion:
./scripts/run-agent-memory-demo.sh full
Promotion is asynchronous; an event write does not make a fact immediately searchable. The sample also displays the durable session, managed summary, and bounded history window separately. See the persistent sample guide for configuration and owner-scoped cleanup.
ASP.NET Core DevUI travel assistant
Combine remembered preferences with current travel bookings through a scoped Context Retriever surface. The sample supports local multi-user sign-in, optional OIDC, and a Compose observability stack.
Generate the model and synthetic records without credentials:
dotnet run --project samples/RedisIrisDevUI -- --print-context-model
dotnet run --project samples/RedisIrisDevUI -- --print-context-seed
Follow the DevUI guide to provision the surface, seed records, configure credentials, and run the UI. Provisioning and seeding change the chosen service; the guide distinguishes those operations from running the client.
Context Retriever POCO model
Export a deterministic DataModel from annotated .NET types without connecting to a service:
dotnet run --project samples/RedisIrisContextModel -- ./travel-context-model.json
The optional --deploy mode uses explicitly supplied administrative credentials and creates a new surface on each run.
See the model sample and modeling guide.
C# notebooks
The notebook guide covers kernel setup, environment files, credential bundles, and all walkthroughs.
| Notebook | Walkthrough |
|---|---|
00 |
Shared Context Retriever provisioning and seeding, with executable PowerShell and Bash variants. |
01 |
Agent Memory data plane: events, summaries, search, and scoped cleanup. |
02 |
Agent Framework memory: durable history and cross-session recall. |
03 |
Context Retriever fundamentals: discovery, queries, and per-invocation tools. |
04 |
POCO modeling: compilation, relationships, comparison, and a typed live read. |
05 |
Remembered preferences combined with current travel records. |
06 |
Two users with isolated context and optional live authorization checks. |
07 |
Tune history windows, summaries, filters, thresholds, and recall budgets. |
08 |
Diagnose failures, cancellation, traces, and metrics. |
Notebooks 06–08 run offline by default with a notebook-only HTTP fixture and need no service or model credentials. Live notebook prerequisites vary; follow the notebook guide before running cells. Clear outputs and execution counts before sharing or committing:
python3 eng/check-notebook-outputs.py