repl

repl

Persistent Python REPL tool — maintains state across calls within a session.

Unlike CodeInterpreterTool (which runs each snippet in a fresh subprocess), this tool keeps variables, functions, and imports alive across invocations within the same session.

Classes

ReplTool

ReplTool(
    timeout: int = 30,
    max_output: int = 10000,
    max_sessions: int = 16,
)

Bases: BaseTool

Persistent Python REPL with session management.

PARAMETER	DESCRIPTION
timeout	Maximum execution time in seconds per call. TYPE: int DEFAULT: 30
max_output	Maximum characters of captured output. TYPE: int DEFAULT: 10000
max_sessions	Maximum concurrent sessions (LRU eviction). TYPE: int DEFAULT: 16

Source code in src/diapason/tools/repl.py

def __init__(
    self,
    timeout: int = 30,
    max_output: int = 10000,
    max_sessions: int = 16,
) -> None:
    self._timeout = timeout
    self._max_output = max_output
    self._max_sessions = max_sessions
    self._sessions: Dict[str, _ReplSession] = {}
    self._lock = threading.Lock()
    self._process_context = multiprocessing.get_context("spawn")

Methods:

close

close() -> None

Release all REPL worker processes.

Source code in src/diapason/tools/repl.py

def close(self) -> None:
    """Release all REPL worker processes."""
    with self._lock:
        sessions = list(self._sessions.values())
        self._sessions.clear()
    for session in sessions:
        self._stop_session(session)