Runner level context compaction ideas

[maestromaximo]

Please read this first

• Have you read the docs?Agents SDK docs
• yes I have
• Have you searched for related issues? Others may have had similar requests
• Yes I have

Enablement of compactation for gpt-5.2 -like models.

Alike to how codex is able to call for compactation, it would be great to be able to enable our agent Runner loop to be able to have the option to do so as well.

I also considered tat the closest thing we could have to compactation open source would be Deepseek OCR, so possibly allow an api endpoint that is openai-like yet allows for other types of future comapctation endpoints?

[ruskaruma]

@seratch quick question on scope and feasibility here.

Enabling compaction in the Agent Runner (similar to Codex-style compaction) feels potentially impactful for long running agents, especially with GPT-5.x class models.

From your perspective, does this align with the SDK’s intended direction? Or would this require core Runner changes, or could it live as an optional / pluggable mechanism? Or is this intentionally out of scope for now?

I am finding this idea quite interesting and would be happy to dig deeper or help prototype if this is something the team would consider. Let me know about your thoughts on this one. Thankyou for your time :-)

[seratch]

The only plan we currently have is to enable responses.compact API for sessions like you ported my TS enhancement: #2224

That said, if there are any simple ideas to unlock compaction for an agent loop, we'd love to hear them.