Strata 0.1.39 adds concurrent requests and Responses API support

Strata adds optional batching and stateless Responses API support

Jordan Reid
October 4, 2026
Tools, AI Agents
1 min

A gaming desktop with a glass side panel shows illuminated cooling fans and components on a computer shop display — Gaming PC photographed by Prolete in 2018. Archival illustration rather than a Strata test system

Listen to this article

Strata released version 0.1.39 on October 4, adding optional concurrent requests and a Responses API endpoint for Codex CLI to its local AI runtime.

More conversations need more memory

Concurrency remains disabled by default. The maintainer reports that the last of four requests on an RTX 5070 started sooner than sequential serving, while combined generation slowed. Each slot reduces memory available for cached experts.

Batched output can differ from solo output under default settings. The batching guide also says repetition, frequency and presence penalties do not apply in batch windows.

Codex support has boundaries

The maintainer tested Codex CLI 0.160.0, including a tool loop. The endpoint is stateless, so clients resend conversation history. Hosted tools, stored response continuation and reasoning summaries are unsupported.

Older GPU, Intel Arc and older CPU paths remain experimental. Some contributor hardware was unavailable to the maintainer.

The README speed table was not measured again for this release.

For a local agent workflow, compare the delay before output starts with the time until every job finishes. Check the client settings your work depends on, using a small trial before replacing a working setup.

ByteForward reviewed release notes and versioned documentation. We have not installed or tested Strata.

Archival gaming PC photograph by Prolete / CC0 / Wikimedia Commons, dated February 2, 2018. Used unchanged as an illustration. This is not a Strata test system.

Jordan Reid

Jordan Reid is focused on AI tools, agents, developer products, and the way technology changes everyday work. Jordan approaches a launch from the user’s side of the screen. What can it actually help someone finish? The voice is practical, conversational, and skeptical of products that turn a simple job into five new settings. Coverage follows coding assistants, creative software, browser agents, and the workflows around them, with attention to pricing, permissions, setup, and the human work that remains.