Skip to content

[FEATURE] OpenAI Responses server-side compaction #5409

Description

@jimbrub

Human Overview

Server-side compaction through OpenAI's API.

SDK Language

Both / Not language-specific

Problem Statement

The OpenAI Responses API compacts server-side and returns an encrypted item with no readable summary. Strands' summarization on that model runs client-side only.

Proposed Solution

Implement compact() for the OpenAI Responses model. Decide whether CompactionBlock can carry both providers' output (0022 open question "Is CompactionBlock provider-neutral?"), and extend #5402 to drop a block with no readable content when switching providers.

Use Case

An agent on an OpenAI reasoning model summarizes without losing its encrypted reasoning.

Additional Context

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

area-contextSession or context relatedarea-modelRelated to models or model providersenhancementNew feature or request

Fields

Language

None yet

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions