Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsCodex memory is not a promise that every conversation is saved and replayed later. The documented Codex repository workflow uses a staged, file-based process to select useful information and consolidate it into reusable memory files. OpenCode’s reviewed documentation instead describes persistent instructions and on-demand skills; it does not describe an equivalent automatic memory-consolidation pipeline.
What does Codex remember between sessions?
In the Codex repository’s documented memory workflow, durable memory lives in files rather than in a guaranteed replay of every previous chat. The repository separates the process that reads memory from the process that writes it: the read path handles instruction injection and citation parsing, while the write path supports prompt rendering, filesystem operations, and workspace diffs. See the OpenAI Codex repository; its main branch is mutable, so these details describe the documented repository workflow rather than a fixed release or every Codex product surface.
As an Amazon Associate I earn from qualifying purchases.
Two files serve different purposes
memory_summary.mdis the compact summary intended to be loaded into the prompt.MEMORY.mdis a searchable, retrieval-oriented handbook. The consolidation template describes it as “the durable, retrieval-oriented handbook.”
The template directs the writing process to surface useful, recently updated, validated information near the top. When a task family is ambiguous, important, or duplicated, it can consult rollout summaries for more evidence. A summary can also lead to opening fuller memory details when they are useful, rather than loading the whole handbook into every prompt. These are design instructions, not a guarantee that a particular item will be saved or retrieved.
Who decides what Codex remembers?
The documented decision-maker is a staged memory-writing and consolidation process guided by its instructions. It is expected to select useful, validated signal and maintain the durable files; the documentation does not identify a human editor who approves every entry or promise complete retention. Users can separately author or edit project instructions such as AGENTS.md, but those instructions are not the same thing as generated memory artifacts.
#1 Best Overall
That distinction matters: a project instruction tells an agent how to work, while a memory file can preserve distilled information for later retrieval. Neither establishes that all past conversation content remains available.
Does Codex remember every conversation?
No such guarantee is established by the documentation. Memory files, conversational session state, and context compaction are distinct mechanisms:
- Session or thread state: the Agents API guide describes saved configuration, turns, and items. It also says the managed Codex harness performs automatic context compaction. Compaction manages the working context; it is not evidence that a full transcript becomes durable memory.
- File-based memory: the Codex repository documents memory files and read/write components. The Agents SDK sandbox guide describes another file-based memory capability and explicitly distinguishes it from conversational session history. That SDK behavior should not be treated as proof of how every Codex product surface works.
- Project instructions:
AGENTS.mdcan provide reusable guidance discovered for a project. It is an instruction source, not inherently a record of previous conversations.
For the separate concepts, see the Agents API guide, the Agents SDK sandbox memory guide, and the Codex prompting guide. Their descriptions apply to their respective API, SDK, or CLI contexts; they should not be collapsed into one universal memory behavior.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →How is Codex memory different from AGENTS.md?
Memory and AGENTS.md can both help across work sessions, but they have different roles. The repository’s memory workflow describes selecting and consolidating useful information into durable, searchable files. AGENTS.md is a source of instructions that Codex can discover and inject for a project. You can inspect and edit an instruction file directly; the documented memory workflow assigns writing and consolidation to its memory process.
The Codex CLI prompting documentation describes instruction discovery and injection. This is a documented behavior of that CLI context, not proof that every Codex interface uses identical discovery rules.
How does OpenCode memory work?
OpenCode’s reviewed documentation describes persistence through instructions and reusable skills rather than through a documented automatic conversation-to-memory pipeline. Its rules documentation describes global and workspace AGENTS.md guidance accumulated by scope. Its skills documentation describes discoverable skills in project or global locations that are loaded on demand through the skill tool.
Those mechanisms can make guidance available in later work, but the pages reviewed do not describe an automatic process that extracts conversation details, validates them, and consolidates them into memory files as the Codex repository workflow does. This is a comparison of the published instructions and skills documentation, not a claim that OpenCode cannot be extended with a plugin or a user-built workflow.
Codex and OpenCode: documented mechanisms compared
| Area | Codex documentation reviewed | OpenCode documentation reviewed |
|---|---|---|
| Persistent project guidance | AGENTS.md discovery and injection are described in the Codex CLI prompting guide. |
Global and workspace AGENTS.md files are documented, with instructions accumulated by scope. |
| Reusable skills | OpenAI documents skills for agent workflows; the Codex memory sources focus on memory files and instruction handling. | Skills can be discovered in project or global locations and loaded on demand through the skill tool. |
| Durable memory pipeline | The Codex repository documents distinct memory read/write paths and consolidated memory files. | The reviewed rules and skills pages do not document an equivalent automatic consolidation pipeline. |
| Session continuity | The Agents API guide describes saved configuration, turns, and items, and automatic context compaction in the managed Codex harness. | The reviewed rules and skills pages do not establish session-persistence behavior. |
What to check when comparing agent memory
“Memory” can mean very different things across tools. To evaluate a setup for your work, check what actually persists and how it becomes available:
Best Value
- Scope: Is information kept in a thread, a project instruction file, or a cross-run memory file?
- Authorship: Do you write and maintain the guidance, or does an agent process generate and consolidate it?
- Retrieval: Is content always injected, summarized in the prompt, or opened only when relevant?
- Inspectability: Can you view and edit the files that shape later behavior?
- Automation: Do the docs describe extracting and consolidating past information, or only loading instructions and skills?
These questions separate durable memory from session continuity and project guidance without assuming that similarly named features behave alike.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
