feat: persist and inherit per-thread multi-agent runtime by jif-oai · Pull Request #25708 · openai/codex

jif-oai · 2026-06-01T22:35:29Z

Why

multi_agent_version was effectively a live feature-flag decision. That meant a thread could switch runtimes when it was resumed, forked, or spawned under different config than the one that created it, and it made model-catalog runtime selectors unreliable if the user changed models before the first real turn. The result was drift in tool surfacing, interrupt history markers, usage-hint filtering, and thread concurrency limits.

What changed

Added MultiAgentVersion to rollout SessionMeta and model-catalog ModelInfo, plus helpers to recover it from stored history.
Resolve the runtime from the selected model on the first real turn, persist it once for non-ephemeral threads, and treat legacy resumed/forked histories with no recorded runtime as V1.
Thread start/resume/fork/spawn and agent-job worker flows now inherit the resolved runtime and use it consistently for tool gating, usage-hint behavior, parent notifications, interrupt markers, and max-thread calculations.
Added ThreadStore::set_multi_agent_version_if_unset so older rollouts can be backfilled without rewriting thread history wholesale.

Testing

Added coverage in core/tests/suite/model_runtime_selectors.rs, app-server/tests/suite/v2/thread_resume.rs, app-server/tests/suite/v2/thread_fork.rs, core/src/session/tests.rs, and thread-store/src/local/mod.rs.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f8aa631734

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Stack split from #25708. Original PR intentionally left open. This first PR adds the multi-agent runtime metadata types and catalog plumbing used by the rest of the stack.

Stack split from #25708. Original PR intentionally left open. This second PR persists multi-agent runtime metadata through thread creation, rollout recording, and thread storage.

Stack split from #25708. Original PR intentionally left open. This third PR resolves the effective per-thread multi-agent runtime from persisted metadata, inherited runtime, and current model selection.

Stack split from #25708. Original PR intentionally left open. This fourth PR adds coverage that remote model multi-agent runtime selectors override local feature flag defaults.

Stack split from #25708. Original PR intentionally left open. This fifth PR adds coverage that a remotely selected multi-agent runtime is applied when the model is selected before the first turn.

Stack split from openai#25708. Original PR intentionally left open. This first PR adds the multi-agent runtime metadata types and catalog plumbing used by the rest of the stack.