OpsfleetAgent Bootcamp

Task 3 · 8 tasks

File sessions

Persist the conversation to disk so the agent still knows Alice after a restart.

15 minEasy
Alice’s ask

The “Goldfish” problem

Alice introduced herself. One restart later the agent greeted her like a stranger. Give it short-term memory that survives the process.

You

Build it

  1. Make the conversation survive a restart

    Challenge

    Give the Task 2 agent a session that is saved under SESSIONS_DIR with the id SESSION_ID (alice_session_starter in your starter, alice_session in the reference, so the reference never fills in your starter's memory).

    Work in phase1/starter/t3_sessions.py (look for TODO; an unfinished one prints [starter] TODO …) and run it with uv run bootcamp.py phase1 t3 --starter. The reference solution is phase1/t3_sessions.py; uv run bootcamp.py phase1 t3 runs it.

    Docs
    Hint 1

    Strands has session managers that plug into the agent lifecycle. Look at strands.session.FileSessionManager. See Strands: session management.

    Hint 2

    Create it once, before the agent, and pass it as session_manager=.

    sketchread only
    session_manager = FileSessionManager(session_id=..., storage_dir=str(SESSIONS_DIR))
    Solution

    Paste each block below into phase1/starter/t3_sessions.py in place of the definition with the same name (the TODO), including any import line that comes with it. Nothing else in the file changes.

    phase1/starter/t3_sessions.py · replace build_session_manager() (keep its import line)
    from strands.session import FileSessionManager


def build_session_manager() -> FileSessionManager:
    """Session SESSION_ID, stored as JSON files under SESSIONS_DIR."""
    return FileSessionManager(session_id=SESSION_ID, storage_dir=str(SESSIONS_DIR))

  2. Introduce yourself, then restart

    terminal · your starter file
    uv run bootcamp.py phase1 t3 --starter "My name is Alice Chen, CEO of DataStream Corp"
uv run bootcamp.py phase1 t3 --starter "What's my name?"
    terminal · reference solution
    uv run bootcamp.py phase1 t3 "My name is Alice Chen, CEO of DataStream Corp"
uv run bootcamp.py phase1 t3 "What's my name?"

    Two separate processes, one memory. Look at what was stored:

    terminal
    ls phase1/sessions/session_alice_session_starter/agents/agent_default/messages/   # your starter
ls phase1/sessions/session_alice_session/agents/agent_default/messages/           # reference

    To start over, pass a new session name: --session alice2 (any letters, digits, _ or -) starts an empty session, and the run prints [session] ... so you always see which one you are in. Deleting a session's folder works too.

  3. Experiments

    • Open a message_*.json file. What exactly is persisted?
    • Play two users: introduce yourself as Bob with --session bob, then ask --session bob "What's my name?" and the same without --session. Now run both people on one shared session name: each sees the other's history. That's the isolation problem AgentCore Memory solves in Phase 2.

Check your work

Phase 1 has no automated test: you check it by running the task and looking for the result below.

In the second process, “What's my name?” answers Alice Chen without a [tool] line: the name comes from the restored history, not the database. JSON message files appear under phase1/sessions/session_alice_session_starter/ (reference: session_alice_session/). If your starter remembers nothing, check that it saves to SESSION_ID.

Failed? Run the same prompt on the reference (drop --starter). If the reference fails too, it's the model: rerun or try another MODEL_ID (uv run bootcamp.py llm models).

Under the hood

FileSessionManager hooks into the agent lifecycle: it restores messages when the agent is created and appends every new message as a JSON file. It's ideal for local work and useless for a fleet of containers, which is why Phase 2 swaps it for AgentCoreMemorySessionManager with almost the same call site.