Memex vs OpenAI dots: Personal Records and Ongoing Work
By Clifford · Product research
Written by Memex Lab. This comparison covers documented capabilities, not a hands-on benchmark. Sources checked on October 11, 2026.
You finish a month of exhibition planning with an approved venue, thirty sketches, and a voice note explaining why you abandoned your first idea. Next month, the agent may only need the venue and deadline. You may still want every sketch and that voice note.
OpenAI dots follows ongoing responsibilities across conversations and tools. Memex starts with personal records on your device and lets you configure the agents and model connections around them. Here, dots means OpenAI’s ChatGPT product, rather than the unrelated open-source projects with similar names.
At a glance
| What matters | Memex | OpenAI dots |
|---|---|---|
| Starting point | Personal notes, photos, voice, and the archive they form | Ongoing responsibilities across connected tools |
| Agent execution | Device app; model inference depends on the endpoint | Cloud agent; can also coordinate authorized local tasks |
| Memory | Original records, cards, knowledge files, and agent memory | ChatGPT memory plus the dot’s separate saved notes |
| Model configuration | Supported provider APIs, per-agent models, optional managed access | Overview identifies GPT-6 Astra; delegated tasks have their own settings |
| Portability | Markdown record export and backup/restore | Save task results; a complete dot-memory export is not established here |
A deadline and the reasons behind it
Imagine preparing an exhibition. An agent can keep the venue discussion moving, track the printing deadline, and prepare replies. Dots is designed to keep following such responsibilities after the conversation ends. That is useful when the work changes as people respond.
Your personal record has a different job: keep the first sketch, the voice note after visiting the space, and the reason you changed the title. In Memex, those originals can sit beside generated cards and knowledge files. You can revisit them even when the next action has already been completed.
Dots also has persistent memory. Its documentation distinguishes relevant ChatGPT memory from the dot’s own notes, and says those notes are not a complete transcript. A remembered preference is useful working context; it should not be your only copy of material you want to preserve.
How dots uses tasks and memory · How Memex organizes personal records
Three locations to keep separate
Memex stores its primary records in local files and SQLite, and runs its agent workflow inside the device app. The model can be elsewhere. A configured cloud provider receives request context; an Ollama model runs wherever its endpoint is hosted. A phone connected to a computer on your network is not doing all inference on the phone.
Dots has its own cloud computer and browser, so cloud work can continue when your devices are off. It can also create Work or Codex tasks on a connected personal computer. That computer must remain online with the ChatGPT app open. Connecting local files does not turn the cloud agent into a local-only service.
For Memex, device battery, operating-system background limits, and endpoint availability matter. Do not choose it on the assumption that a phone-local workflow will behave like an always-on hosted computer.
Dots cloud and connected computers · What local-first storage means
Privacy means checking the actual data path
Dots applies ChatGPT data controls to eligible conversations and work. OpenAI says it does not train models directly on proactive research or private notes; information incorporated into an eligible conversation or task follows your data settings. Cloud processing and training use are separate questions.
Memex gives you a local starting point and a configurable inference path. Direct provider configuration sends requests to that provider. Optional Memex AI access proxies model calls. Sync, backups, transcription, and external tools can create additional data paths, so check the configuration you actually use.
Dots provides automatic action review, app permissions, and optional custom rules. If the exhibition agent drafts a venue reply, reviewing where the information goes matters as much as reviewing the prose. With any agent, connect only the sources needed for the job.
Dots data settings and action controls · Privacy checks for personal records
Change the model without starting your archive again
Memex’s source is GPL-3.0. You can inspect it, build it, or modify it under the license. In the app, open the avatar menu and Model Configuration to choose a supported provider and connection method. Different agents can use different models; custom prompts, Skills, and working directories shape their tasks.
For the exhibition, configure a reflection agent with a narrow instruction: “Use only these dated notes. List changes of direction, cite the source record for each, and leave uncertain connections as questions.” Run it on a small folder first. Then change its supported model and compare the same records. Check whether a detail was invented, whether a citation opens the right original, and whether your phrasing survived the summary.
The dots overview identifies GPT-6 Astra. Its delegated Work and Codex tasks are separate execution contexts; the documentation reviewed here does not establish an arbitrary provider-API setting for the dot itself. Check task-level settings separately rather than assuming every part uses one model.
Memex provider configuration · OpenAI dots overview · Planning a local model connection
Test the exit and the cost before expanding
Memex documents Markdown record export and full backup/restore. Test both with a small sample: open the exported text in another editor and check which attachments and metadata you still need. A readable export and a restorable application backup solve different problems.
For dots, save the task outputs you need before deleting the dot. Its deletion guide tells you to read the confirmation for the treatment of conversations, memory, and schedules. The reviewed docs do not establish a complete portable export of the dot’s notes and state; that is a question to verify in your account, not evidence that results cannot be saved.
Access to dots is rolling out by eligible plan, region, age, and workspace policy. Delegated Work and Codex tasks use their respective limits. Check your account before paying for access. For Memex, estimate the chosen provider, model, context size, and number of runs rather than assuming open source means zero running cost.
Memex export and backup documentation · Save results before deleting a dot · Dots access and usage conditions · Compare all five personal agents
Start with what you want to keep
If the recurring problem is following through with venues and collaborators, try dots on that responsibility. If the problem is losing the material behind your decisions, start with Memex and test the archive before adding automation.
Keep one exported note open beside the agent’s review. You should be able to tell which words are yours and which conclusions the model added.