What should OpenAI teams do about it?

Review interrupt_agent semantics in your multi-agent orchestration code - it no longer destroys agent state • Test web search integration in code mode before rolling out to production tasks • Verify session residency behavior under load - concurrency admission is now best-effort

MULTI-AGENT V2 RESIDENCY LAYER COMPLETE

By RepoJournal · Filed 22:42 UTC on June 8, 2026 · About OpenAI

Codex shipped the memory management backbone that separates durable agent identity from thread residency, unblocking true concurrent execution counting across the multi-agent stack.

The v2 agent model just matured past the threading bottleneck. After last week's reload-on-delivery change, agents could survive thread unloads; today's LRU paging [1] moves that separation into `ThreadManager` itself, letting the system page out loaded subagents when sessions get crowded without losing logical identity. That foundation enables the concurrency recount [3] shipping in parallel: v2 now counts active non-root turns instead of resident threads, with admission checks that stay synchronous but allow slight overshot on competing checks. The API surface gets sharper too [4]. Renaming `close_agent` to `interrupt_agent` [4] fixes the contract - v2 agents stay reusable after interruption, so the tool name should match what it does: pause the current turn, not destroy the agent. Web search in code mode [2] also lands today, consuming plaintext output from `/v1/alpha/search` and exposing `web.run` to nested JavaScript calls. Finally, the resume path gets fixed [5]: the old logic still eagerly reopened entire descendant trees, which defeats the point of residency management. Interrupted agents that go idle are now eligible for eviction instead of staying pinned.

FAQ

What changed in OpenAI on June 8, 2026?: Codex shipped the memory management backbone that separates durable agent identity from thread residency, unblocking true concurrent execution counting across the multi-agent stack.
What should OpenAI teams do about it?: Review interrupt_agent semantics in your multi-agent orchestration code - it no longer destroys agent state • Test web search integration in code mode before rolling out to production tasks • Verify session residency behavior under load - concurrency admission is now best-effort
Which OpenAI repositories shipped on June 8, 2026?: openai/codex

MULTI-AGENT V2 RESIDENCY LAYER COMPLETE

The showcase is a teaser. Your wire is the product.

The showcase is a teaser.
Your wire is the product.