Thanks — I see you're trying to run multiple conversations in parallel and hitting

RuntimeError: cannot start agent: an activity is already running

. I’ll help you get parallel test runs working. Short answer • That error happens because an

AgentSession

(and related runtime resources) isn’t designed to have multiple concurrent activities in the same process/context. To run tests in parallel, isolate each

AgentSession

so no two run in the same process or share the same runtime objects. What you can do (practical options) • Run each session in its own process ◦ Use Python’s

multiprocessing

or run tests across separate processes. This ensures each

AgentSession

has its own runtime and avoids the “activity already running” conflict. • Use a test runner that distributes work across processes ◦ Use

pytest-xdist

to run test files or test functions in separate worker processes. Each worker will create its own

AgentSession

, avoiding shared activity conflicts. • Create process-isolated workers manually ◦ Start a small pool of worker processes where each worker creates and owns its

AgentSession

. Send test scenarios to workers and collect results. • If you must stay in a single process, ensure strict sequential usage ◦ Only start one

AgentSession

activity at a time (not parallel). Use sequential

async with AgentSession(...)

blocks or an execution queue to avoid overlap. • Ensure you’re not sharing server/job resources ◦ Don’t share

JobProcess

,

AgentServer

, or any runtime objects between tests. Each parallel test should initialize its own resources if those objects are not concurrency-safe. Async-specific notes • If using asyncio test frameworks (e.g.,

pytest-asyncio

), ensure each concurrent test uses separate event loop contexts or run them in separate processes. Running multiple sessions on the same event loop can lead to conflicts if the agent runtime expects exclusive control. Example approaches 1) Quick multiprocessing sketch

from multiprocessing import Process

def run_session(test_input):
    async def _run():
        async with AgentSession(...) as session:
            return await session.run(user_input=test_input)
    asyncio.run(_run())

procs = []
for inp in inputs:
    p = Process(target=run_session, args=(inp,))
    p.start()
    procs.append(p)
for p in procs:
    p.join()

2) Use pytest-xdist for test suites • Install

pytest-xdist

and run tests with

pytest -n auto

(or a fixed number). Each pytest worker runs in its own process, so each

AgentSession

is isolated. Questions to narrow it down • Are you running tests with pytest (and

pytest-asyncio

) or a custom runner? • Do you need true parallelism (multiple processes) or just concurrency on a single machine? • Are you currently sharing any objects like an

AgentServer

or

JobProcess

across tests? If you share which runner you’re using and a short snippet of how you create/start

AgentSession

, I can suggest the most direct code changes to run your sessions in parallel. Sources: Testing and evaluation | LiveKit Documentation | Job lifecycle | LiveKit Documentation