标签：#opencode

I began this investigation with a specific question: can Codex autonomously add, modify, and delete its own memories? The product documentation already says that it has memory. What I wanted to know was who actually decides what survives. When an old chat contains a useful build command, does deterministic application code copy it into a database? Does the active coding model call a memory tool? Does another model summarize the chat later? When the command becomes obsolete, is the old fact overwritten, invalidated, aged out, or simply left where future agents may still find it? Those questions led to a more interesting result than a feature checklist. Codex has a genuine cross-session memory subsystem, but its behavior is split between model judgment and deterministic lifecycle code. Models decide what a rollout means and how durable guidance should be rewritten. Runtime code decides which rollouts are eligible, which evidence remains in the working set, when old records are deleted, and when the consolidation model is allowed to run. That makes the short answer precise: With local memories enabled, Codex can autonomously add, modify, merge, and remove persistent memory without a user approving each write. User-requested corrections follow a separate append-only note path, while retention, thread deletion, and reset provide additional forms of forgetting. The rest of this article explains why each word in that answer matters. This analysis is pinned to OpenAI Codex commit 8444cf63b50a8a88521e0d2970d49f659b48eac7 , checked on August 25, 2026. The feature is marked stable in that source tree but remains off by default, so this describes implemented behavior, not behavior every Codex user is currently receiving. Key Takeaways Codex local memory is a background two-model pipeline. One model extracts reusable material from each eligible rollout. A second model consolidates those outputs into a global file-based memory workspace. The LLM owns semantic CRUD, but not lifecy

I started this investigation after finding a local OpenCode project that appeared to remember guidance across sessions. The guidance lived in Markdown files outside the repository, yet every new session followed it. The operational question was simple: had OpenCode decided to preserve those facts, or had someone explicitly written them? The session record answered it. An agent had created the files through an ordinary file tool after an explicit user request. A project-local instructions configuration then loaded them on every provider turn. What looked like autonomous memory was user-triggered file authoring plus deterministic prompt injection. That result sent me looking for the actual memory subsystem. There is no general runtime-managed service that decides what to save, updates facts when they change, and semantically retrieves useful knowledge in later sessions. What users experience as "memory" is produced by three different mechanisms with different owners and failure modes: instruction files are loaded into the system prompt, durable session events and projected messages are persisted in SQLite, and old model-visible context is replaced by a generated compaction checkpoint when the request grows too large. These mechanisms work together, but they do not form an autonomous long-term memory manager. That distinction matters. If a coding agent remembers a project rule because AGENTS.md is injected on every turn, that is not learned memory. If it can reopen an old transcript from SQLite, that does not mean a new session can retrieve facts from it. If a long session survives by summarizing its history, that does not mean the runtime selected the most important information. This article follows the current OpenCode source tree, which contains both the desktop-compatible session path under packages/opencode and the newer V2 runtime under packages/core . Where the two paths differ, I call out the difference rather than treating them as one implementation. Primary c

I have a confession: I roll my eyes at AI benchmarks. Every other week someone on Twitter posts a chart where a brand new model is suddenly beating Opus and GPT, the replies go crazy, and then you actually use the thing and it falls apart on the first real task. Beautiful numbers, ugly code. So when z.ai shipped GLM 5.2 and the timeline started shouting that an open-weights model was now nipping at the heels of the frontier labs, my instinct was the usual one. Sure. Prove it. This post is me proving it. I gave GLM 5.2 a real feature to build on my actual production website, with almost no hand-holding, and watched what happened. Spoiler: I was not expecting to write the sentence I ended up writing. If you'd rather watch me run the whole thing live (including the part where my tools crashed on camera), the video is right here: The claims I was here to test Let's get the hype out of the way first, because the claims are genuinely big. // Detect dark theme var iframe = document.getElementById('tweet-2066938937344495629-390'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=2066938937344495629&theme=dark" } GLM 5.2 is the same physical size as GLM 5.1 (744B total parameters, 40B active), but on the Artificial Analysis Intelligence Index it jumped 11 points, from 40 to 51. That score makes it the leading open-weights model , ahead of MiniMax-M3 (44), DeepSeek V4 Pro (44) and Kimi K2.6 (43). On the overall leaderboard it sits behind only Claude Fable 5 (60), Claude Opus 4.8 (56) and GPT-5.5 (55). For an open, MIT-licensed model you can download the weights for, that is a wild place to be. Here is the upgrade at a glance: GLM 5.1 GLM 5.2 Intelligence Index (v4.1) 40 51 Context window 200K 1M Total / active params 744B / 40B 744B / 40B Output tokens per task 26k 43k Cost per task ~$0.25 ~$0.46 Price (in / cache / out per 1M) $1.4 / $0.26 / $4.4 $1.4 / $0.26 / $4.4 License MIT MIT Two things jump out. First