Skip to content

Shell Sentinel 1.3.0 — OpenAI Responses API

Latest

Choose a tag to compare

@ibitato ibitato released this 27 Jul 10:37

Highlights

  • Adds selectable OpenAI Responses API support with function tools and reasoning, while preserving Chat Completions as the compatibility default.
  • Automatically normalizes token and reasoning parameters for each OpenAI endpoint.
  • Adds configurable stateful Responses chaining and secure MCP environment passthrough.
  • Aligns the English, Spanish and German documentation and static website.

Configuration

Use api = responses to opt in. The example configuration uses stateful = false, reasoning_effort = medium and a 32,768-token output limit. Keep API keys in environment variables. LM Studio continues to use its OpenAI-compatible Chat Completions mode.

Dependencies

  • Strands Agents 1.50.1
  • Strands Agents Tools 0.8.5
  • OpenAI 2.48.0

Validation

  • 143 offline tests passed
  • 2 opt-in OpenAI Responses end-to-end tests passed
  • Ruff, package metadata, dependency integrity, TUI startup and static-site checks passed

Known limitation

Responses reasoning blocks do not yet preserve complete reasoningContent continuity when reconstructing multi-turn conversations; stateful mode is available for controlled evaluation.

See CHANGELOG.md for the complete release notes.