Chronicle gives Codex a memory of your screen — but only on macOS, only for ChatGPT Pro, only for Codex, and it ships your screenshots to OpenAI’s servers to build them. LUCI matches every Chronicle capability and keeps it 100% on-device — then opens that memory to any agent over MCP: Claude, Hermes, OpenClaw, or Cursor.
Life memory for every AI agent — your whole day on the computer, stored on your device. Read it from Claude, Cursor, Codex and the rest, on Windows and Mac.
Screen-only memory for Codex. macOS + ChatGPT Pro only. Screenshots processed on OpenAI servers. Opt-in research preview.
OpenAI’s Chronicle is a smart idea locked in a small box: it watches your screen, builds markdown memories, and feeds them back to Codex so you restate less context. The memory design is genuinely clever — a two-stage pipeline where an extract model pulls candidate memories from each thread and a consolidation model merges them globally in the background, keeping only durable, reusable context: stable preferences, recurring workflows, your tech stack, project conventions, known pitfalls. When a better source of truth exists — a file, a Slack thread, a Google Doc, a PR — it stores a pointer to that source rather than the content. But that design has a flip side: Chronicle is an index of your work, not a record of it. Anything its summarizer judges one-off is dropped at extraction time and gone for good. Then come the fences: it’s an opt-in research preview limited to macOS and ChatGPT Pro, it captures screenshots only — no meetings, no mic, no system audio — it uploads frames to OpenAI’s servers to generate the memories, and the result is readable by exactly one agent: Codex. LUCI does the same distilled-memory job without the fences, and keeps the record underneath the index. Everything is processed and stored 100% on your own device — nothing is uploaded — meetings and audio are captured alongside your screen, and the memory is exposed as an MCP tool any agent can call. Point Claude, Hermes, OpenClaw, or Cursor at your LUCI memory and they get the same 'knows what you were doing' context Chronicle gives Codex — on Windows and Mac, with no research-preview waitlist and no rate-limit tax.
共 11 项，LUCI 领先 9 项。
| 功能 | LUCI | Codex Chronicle |
|---|---|---|
| 100% local — no screenshots uploaded | 支持All processing on-device | 不支持Screenshots processed on OpenAI servers |
| Works with any AI agent (MCP) | 支持Claude, Hermes, OpenClaw, Cursor… | 不支持Codex / ChatGPT only |
| Cross-platform (Windows + Mac) | 支持 | 不支持macOS only |
| Captures meetings + audio | 支持Mic + system audio | 不支持No mic / no system audio |
| Screen memory | 支持 | 支持 |
| Searchable record behind the memory | 支持Drill from summary to the exact moment | 不支持Durable summaries + source pointers only; one-off details discarded |
| Encrypted local storage | 支持 | 不支持Unencrypted markdown on disk |
| No aggressive rate limits | 支持 | 不支持Consumes rate limits quickly |
| Generally available (not a preview) | 支持 | 不支持Opt-in research preview |
| No paid-tier lock-in | 支持Standalone app, free | 不支持Requires ChatGPT Pro |
| Pricing | Free — 100% local | Bundled in ChatGPT Pro ($200/mo) |
Standalone app on Windows & Mac. 100% local. Free forever. Memory usable by any MCP agent.
macOS only. Screenshots processed on OpenAI servers. Memory usable only by Codex.
No research-preview gate and no ChatGPT Pro requirement. Grant screen recording and mic permissions to enable screen + meeting memory.
LUCI processes screen and audio into memory locally — unlike Chronicle, no screenshots are uploaded to a server to build your memories.
Point Claude, Hermes, OpenClaw, or Cursor at LUCI’s MCP server. Any agent can now query the same local memory Chronicle reserved for Codex.
Turn on meeting capture for audio and transcripts — memory of conversations and decisions, not just what was on screen.
在本地捕获每一场 Zoom、Meet 和 Teams 会议——没有机器人加入通话——并把它变成可搜索的笔记和待办事项。
Chronicle captures your screen locally, then uploads selected frames, OCR text, and file paths to OpenAI’s servers to generate memories. LUCI does the entire capture-to-memory pipeline on-device — nothing is uploaded, so your screen contents, meetings, and documents never touch someone else’s server.
Chronicle memory is only usable by Codex inside ChatGPT. LUCI exposes your memory as an MCP tool, so any agent can query it — Claude, Hermes, OpenClaw, Cursor, or your own. You’re not locked into one vendor’s assistant to get value from your own context.
Chronicle explicitly has no access to your microphone or system audio — it can’t remember what was said in a call. LUCI captures meetings (Zoom, Meet, Teams, in-room audio) alongside your screen, so your memory covers conversations and decisions, not just what was visible.
Chronicle’s memory is deliberately lossy: it extracts only what looks durable (preferences, workflows, stack, conventions, pitfalls), stores pointers to files, Slack threads, and docs instead of their content, and consolidates everything in a background pass. Elegant — until the detail you need was judged one-off and dropped, or the pointer targets a thread you can no longer find. LUCI builds the same kind of distilled, agent-ready memory and keeps the searchable capture underneath it, so you can always drill from the summary to the actual moment on screen.
Chronicle is a macOS-only, ChatGPT-Pro-only research preview that burns rate limits generating memories in the background. LUCI ships today on Windows and Mac, with no background agents eating your quota and no research-preview gate.
Chronicle is an opt-in research preview from OpenAI that augments Codex memories with context from your screen. It periodically captures screenshots, uses Codex to summarize them into markdown memories, and feeds that context back so you restate less when prompting Codex. Under the hood it runs two stages — an extract model that pulls candidate memories from each thread, and a consolidation model that merges them into global memory during idle time — storing summaries, durable entries, recent inputs, and supporting evidence as local markdown. It’s currently limited to macOS and ChatGPT Pro subscribers.
Chronicle is deliberately selective. Per OpenAI’s docs it keeps durable, reusable context — stable preferences, recurring workflows, your tech stack, project conventions, and known pitfalls — and when a better source exists (a file, Slack thread, Google Doc, dashboard, or PR) it stores a pointer to that source rather than the content itself. Anything judged one-off never makes it into memory. That’s smart for prompting Codex, but it makes Chronicle an index, not a record: if a detail was filtered out at extraction time, there’s nothing to go back to. LUCI keeps both layers — distilled memory for agents, plus a searchable local record of your screen and meetings — so the exact moment is still there when you need it.
Partially. Screen captures are stored temporarily on your device and the final memories are stored locally as unencrypted markdown — but to generate those memories Chronicle uploads selected screenshot frames, OCR text, timing, and file paths to OpenAI’s servers for processing. LUCI, by contrast, runs the entire pipeline on-device: nothing is uploaded.
No. Chronicle memory is only accessible to Codex within the ChatGPT app. LUCI exposes memory over the Model Context Protocol (MCP), so Claude, Hermes, OpenClaw, Cursor, and any other MCP-compatible agent can query the same local memory.
No. Chronicle is only available on macOS, and only for ChatGPT Pro subscribers. LUCI runs on both Windows and macOS.
No. Chronicle captures screen images only — it explicitly has no access to your microphone or system audio, so it can’t remember what was said in a call. LUCI captures meetings and audio (Zoom, Meet, Teams, in-room mic) in addition to your screen.
Mac 和 Windows 上均免费，完全本地运行。