2026.10.10
Memex: Turning Life Fragments into a Local Knowledge Base
Memex is an open-source, local-first AI journal for iOS and Android, aimed at people who want low-friction capture while retaining control over storage and export.
What it does
Memex is not designed primarily for polished daily entries. It accepts fragments such as text, photos, and voice, then uses a multi-agent system to organize them into timeline cards, knowledge records, and insights. The result is closer to a continuously growing personal knowledge base than a conventional daily journaling app.
The project targets iOS and Android and is released under GNU GPL v3.0. That matters to users who want to inspect, modify, or preserve a migration path through open source. The project’s license and privacy materials should still be reviewed before adoption.
Image source: memex. Published by the project; the interface may vary by version.
Known capabilities
According to the README, Memex can produce cards for tasks, routines, events, progress, articles, snippets, quotes, links, conversations, people, places, metrics, ratings, transactions, specification sheets, and galleries. It also lists automatic tagging, entity extraction, cross-reference links, and a conversational assistant for discussing cards or topics.
For knowledge organization, Memex uses a P.A.R.A.-based structure. Its insight cards include charts, summaries, comparisons, maps, routes, timelines, and galleries. Users can also create AI characters with different personalities, receive automatic commentary, chat one-to-one, and import SillyTavern character cards.
The README says records, cards, and knowledge remain on the device, using filesystem storage and SQLite. It lists iCloud Drive, device storage, and app storage as storage options, along with full backup and restore, Markdown export, and biometric app locking. Local-first does not mean every operation is offline: prompts are still sent to the model provider selected by the user. The documented options include OpenAI, Claude, Gemini, DeepSeek, Ollama, and OpenRouter. Some OAuth integrations are explicitly described as unofficial, so they deserve additional scrutiny.
Who it may suit
Memex may suit people who want to capture ideas, photos, voice notes, expenses, trips, or reading fragments quickly and let AI help organize them later. It is also relevant to users who care about device-local storage, want a Markdown exit path, or want to experiment with custom agents from a phone. It may be a poor fit for someone seeking a simple plain-text diary, refusing to configure a model account, or requiring fully offline inference.
Limits and deployment considerations
The first consideration is model dependence. AI features require a configured provider, API key, account, OAuth login, or local Ollama setup. Users must evaluate provider costs, quotas, regional availability, and data practices themselves; the supplied materials do not confirm unified pricing, performance, or minimum hardware requirements.
Second, keeping journal data on the device does not remove the exposure created when prompts are sent to an external model. Private photos, voice recordings, location context, and information about other people require careful provider and integration review. Location context is optional; the README says device GPS and reverse geocoding can attach city, district, or neighborhood information to conversations.
Finally, the feature surface is broad. Custom agents can use skills, work directories, JavaScript execution, external requests, and event triggers. That makes skill files, prompts, and automation settings worth reviewing before allowing access to sensitive material. The v1.0.37 release mentions UI-linked cards in the Agent conversation page and several performance optimizations, but that does not establish stability on every device. Before wider deployment, test backup, restore, export, model configuration, and multi-device workflows with non-sensitive data.
Sources
Sources checked: 2026-10-10