Spotlight: OpenAI Codex 0.145.0 Experimental Paginated Thread History

OpenAI Codex’s 0.145.0 release introduces something that, frankly, I’ve been waiting on for years: paginated thread history with persistent names, search, sub-agent support, and “memories.” This isn’t just a quality-of-life upgrade. It’s a fundamental shift in how we interact with agent sessions, especially if you’re running complex workflows that span days or weeks. The release landed July 21, 2026, and it’s not exaggerating to say it drops Codex into a new league for anyone who’s tired of brittle, transient chats. If you’ve ever cursed at a lost thread or struggled to resume a task from last week, this release will make your life easier. It’s rare to see agent UX improvements this foundational, and I’d argue it matters more than any patch or plugin from the week. Let’s look at what changed, why it’s a big deal, and whether competitors are even close.

What shipped

OpenAI Codex 0.145.0 rolled out a set of experimental features centered around thread history and session management. Here’s what landed:

• Paginated thread history: You can now browse past conversations in a paginated UI. No more endless scrolling or losing context when your chat gets long.
• Efficient resume: Codex lets you jump directly back into a thread, picking up where you left off. This isn’t just for your last session,any thread can be resumed.
• Search: You can search across your conversation history. This applies to individual threads and, crucially, across all your threads.
• Persisted names: Threads can be named and those names stick around. Goodbye generic “Session 23.” Hello “Refactor API endpoints.”
• Sub-agent support: Threads aren’t limited to your main agent. You can branch out, work with multiple sub-agents, and keep their histories tidy.
• Memories: Codex lets you attach “memories” to threads. These are essentially persistent context snippets or instructions that carry over between sessions.

The official changelog calls this “experimental,” but even in alpha, it’s a full-featured session manager. The APIs are already available for developers, and the UI is rolling out to hosted Codex as well as self-hosted instances. If you’re on version 0.145.0 or newer, you get all of this. The changes span issues #33364, #33907, #34085, #34229, and #34029, which tells you this was a multi-week effort. Codex’s team didn’t just slap a search bar on top,they rebuilt core session logic.

Why it matters

If you’re serious about agents, you know session management is a pain. Codex sessions have historically been ephemeral and tied to a single browser window, which makes anything beyond one-off tasks annoying. Now, Codex is introducing a concrete, durable history system that finally treats agent sessions as first-class citizens.

Let’s break down the implications. First, paginated history means you can work on multi-step projects without losing context. Whether you’re debugging code, writing documentation, or orchestrating multiple agents, you can actually go back and see what happened days ago. Before this, you’d have to scrape logs or rely on fragile browser storage.

Efficient resume is a productivity booster. Imagine you’re switching between half a dozen projects, each with its own Codex thread. Now, you can jump back in without re-explaining everything to your agent. This cuts down on redundant prompts and helps maintain continuity.

Search is a game-changer for anyone who’s had to dig through old chats. Codex’s search isn’t just for single threads,it covers all your session history. You can find that obscure error message from last Tuesday or pull up a code snippet without scrolling for minutes. This is the kind of feature that makes agents usable at scale.

Persisted names sound minor, but they’re huge for organization. If you’re running dozens of threads, naming them is the only way to keep track. Codex’s approach is sticky: names persist even if you close the window or switch devices.

Sub-agent support is the sleeper hit. Codex lets you branch threads for different agents, so you can keep your code generation separate from your planning discussions. This is the first step towards true agent orchestration, where you can run multiple specialized agents in parallel and maintain their histories.

“Memories” are the most experimental and maybe the most intriguing. Codex lets you attach persistent snippets to a thread,think recurring instructions, context notes, or links. This is a primitive form of agent memory, and while it’s not full-on long-term memory yet, it’s enough to make multi-session workflows practical.

Taken together, these features push Codex from “chatbot with APIs” to “agent IDE.” You can finally treat conversation history as a real asset, not something you hope survives browser refreshes.

How it compares

Let’s talk competitors. The closest analog is probably Claude Code, which has been inching towards persistent session management but still lags behind. Claude Code’s v2.1.217 (released July 21, 2026) added emoji shortcode autocomplete and a “sandbox.filesystem.disabled” setting, but nothing on the scale of Codex’s thread history overhaul. Claude’s session persistence is basic: you get a single chat history per agent, and search is limited.

Google Gemini CLI, meanwhile, focuses on workspace trust and task isolation. The July 22 release (v0.52.0-nightly.20260722) tightened up RCE protections, which matter for security, but Gemini’s session handling is still primitive. You get logs, but not paginated threads, persistent naming, or multi-agent orchestration. Search in Gemini is rudimentary and limited to the CLI output.

Cline is the newest desktop contender. The July 22 update (v0.0.3) tweaked the UI for agent “Thinking,” but Cline’s session management is still a single-thread affair. No multi-thread, no search, no memories.

None of these competitors offer the full package Codex just shipped. Paginated history, persistent names, search, sub-agent support, and memories,all in one release. Codex is setting the bar for agent session UX, and everyone else is still playing catch-up. The only framework that comes close is LangChain, which supports “reasoning effort” parameters and some session tracking, but doesn’t have a native history UI. LangChain is more developer tool than end-user agent IDE.

If you’re evaluating agent platforms for multi-step projects or orchestrating multiple agents, Codex is now the clear leader. The gap is real.

What to do about it

If you’re on OpenAI Codex, upgrade to 0.145.0 immediately. The new session features aren’t just nice-to-haves. They’re essential for anyone who runs agents beyond one-off tasks. Here’s how to take advantage:

First, start naming your threads. Don’t let “Session 12” become “Session 37.” Use meaningful names so you can pull up specific projects quickly. Codex’s UI now supports this, and the APIs expose thread naming for automation.

Second, try branching threads for sub-agents. If you’re running code generation, planning, and feedback loops, split them out. Codex’s sub-agent support keeps their histories isolated, so you don’t have to worry about cross-contamination of context.

Third, use memories to attach persistent instructions. If you find yourself repeatedly pasting context or reminders, save them as thread memories. This cuts down on prompt engineering and keeps your agents focused.

Fourth, experiment with search. Codex’s search works across all your threads, not just the current one. Use it to find old code snippets, error messages, or decisions. Don’t waste time scrolling,search is fast and accurate.

If you’re building agent workflows for your team, the APIs are now stable enough to integrate into dashboards or orchestration tools. Codex exposes thread management endpoints, so you can automate thread creation, naming, and search. This is a productivity multiplier for anyone building agent-driven systems.

If you’re evaluating competitors, look closely at Codex’s new features. Unless you’re locked into Gemini, Claude, or Cline for some reason, Codex is now the superior choice for session management. The only caveat: these features are labeled “experimental.” Expect some rough edges, but in practice, they’re already stable.

Finally, give feedback. Codex’s team is actively iterating on session UX, and user feedback will determine what sticks. If you run into bugs, report them. If you have feature requests (like long-term memory or cross-thread linking), now’s the time to push.

Bottom line

OpenAI Codex 0.145.0’s thread history overhaul is the most important agent release of the week. It transforms session management from an afterthought to a core feature, making multi-session workflows practical, organized, and efficient. If you care about agent productivity, upgrade now and start using paginated history, search, sub-agent threads, and memories. The competition isn’t close. Codex is now the agent IDE to beat.