Module: SwarmSDK::Defaults::Timeouts

Defined in:

lib/swarm_sdk/defaults.rb

Overview

Timeout values for various operations

All timeouts are in seconds unless explicitly marked as milliseconds. Timeouts balance responsiveness with allowing enough time for operations to complete successfully.

Constant Summary

AGENT_REQUEST_SECONDS =

LLM API request timeout (seconds)

Default timeout for Claude/GPT API calls. 5 minutes accommodates reasoning models (o1, Claude with extended thinking) which can take longer to process complex queries.

300

BASH_COMMAND_MS =

Bash command execution timeout (milliseconds)

Default timeout for shell commands. 2 minutes balances allowing build/test commands while preventing runaway processes.

120_000

BASH_COMMAND_MAX_MS =

Maximum Bash command timeout (milliseconds)

Hard upper limit for bash commands. 10 minutes prevents indefinitely running commands while allowing long builds/tests.

600_000

WEB_FETCH_SECONDS =

Web fetch timeout (seconds)

Timeout for HTTP requests in WebFetch tool. 30 seconds is standard for web requests, allowing slow servers while timing out unresponsive ones.

30

HOOK_SHELL_SECONDS =

Shell hook executor timeout (seconds)

Default timeout for hook shell commands. 60 seconds allows complex pre/post hooks while preventing indefinite blocking.

60

TRANSFORMER_COMMAND_SECONDS =

Workflow transformer command timeout (seconds)

Timeout for input/output transformer bash commands. 60 seconds allows data transformation operations while preventing stalls.

60

EXECUTION_TIMEOUT_SECONDS =

Execution timeout (seconds)

Maximum wall-clock time for entire swarm.execute() call. 30 minutes allows complex multi-agent workflows while preventing runaway execution.

1800

TURN_TIMEOUT_SECONDS =

Turn timeout (seconds)

Maximum time for a single agent.ask() call, including all LLM requests and tool executions. 30 minutes accommodates extended thinking models and complex tool chains.

1800

RESPONSES_API_TTL_SECONDS =

OpenAI responses API ID TTL (seconds)

Time-to-live for cached response IDs. 5 minutes allows conversation continuity while preventing stale cache issues.

300

MCP_REQUEST_SECONDS =

MCP client request timeout (seconds)

Default timeout for MCP server connections. 5 minutes accommodates long-running SSE streams and tool executions. This timeout applies to the entire operation (operation_timeout in HTTPX), so it must be long enough for SSE connections that may run for extended periods.

300