Changelog
v0.2.0 — Provider Wrappers, Automatic Retry, Per-Request Timeouts
April 2026
Provider Wrappers (Python, TypeScript, Go) Drop-in replacements for OpenAI, Anthropic, Google, xAI, and Meta SDKs. Change one import line — every AI call is automatically governed with permits, budget enforcement, usage reporting, and audit trails.
# Before
from openai import OpenAI
# After — same interface, full governance
from keel_sdk.providers.openai import OpenAIAvailable for all five providers across Python, TypeScript, and Go.
Automatic Retry with Exponential Backoff All SDKs now include built-in retry with configurable backoff for transient failures (408, 429, 500, 502, 503, 504). Respects Retry-After headers. Configurable via RetryConfig.
Per-Request Timeouts Override the global timeout on any individual request. Useful for mixing fast permit checks with long-running AI generations.
Governed Execution
Provider wrappers request permits automatically before proxying provider-shaped calls, and the direct execution clients enforce the current execution contract without requiring a caller-supplied permit_id.
Error Classification KeelError now includes is_retryable and retry_after properties, making it easy to build robust error handling.
v0.1.0 — Initial Release
March 2026
Core SDK clients for Python, TypeScript, and Go. Direct access to Keel’s permit, execution, proxy, and job APIs. Streaming support via Server-Sent Events. Express and Next.js middleware for TypeScript.