Context And Memory

What it demonstrates

A ContextProvider gives you two keyword-only hooks around every run:

• before_run(*, agent, session, context, state) — inject instructions, messages, or tools
• after_run(*, agent, session, context, state) — observe the response, update memory

On the context (a SessionContext) you inject via context.extend_instructions(source_id, text), context.extend_messages(...), and context.extend_tools(source_id, tools). History providers are just context providers that persist the transcript: InMemoryHistoryProvider() (default, lost on exit) or FileHistoryProvider("./sessions") (one JSON-Lines file per session, survives restarts).

The code

class ToneAndTally(ContextProvider):
    def __init__(self) -> None:
        super().__init__(source_id="tone_and_tally")

    async def before_run(self, *, agent, session, context, state) -> None:
        context.extend_instructions(self.source_id, "Always answer in British English, one sentence.")

    async def after_run(self, *, agent, session, context, state) -> None:
        state["turns"] = state.get("turns", 0) + 1
        print(f"[provider] turns so far this session: {state['turns']}")

Both a FileHistoryProvider("./sessions") and this ToneAndTally() go into Agent(..., context_providers=[...]) — providers stack, and the file history persists the transcript while ToneAndTally shapes style.

The gotcha

Use the state dict, not instance attributes, when the count should travel with the session — the framework threads that per-provider dict through the session. Store the turn counter on self and it won’t follow the session across reuse.

Azure / MAF mapping

The agent runs on FoundryChatClient (Azure AI Foundry) with AzureCliCredential. Context providers live in the unified context_providers=[...] list — history and custom context alike — independent of the chat client.

Run it

uv run tutorial/02-agents/03_context_and_memory.py — needs Foundry creds. It worked if you get two British-English answers with a turn counter and a persisted transcript under ./sessions/.

Next: Observability