# Memex: a local-first AI journal for iOS and Android that files your fragments for you > Memex is an open-source Flutter app that captures text, photos and voice, then runs a multi-agent system on your device to turn them into structured cards. It is GPL-3.0, it brings its own LLM, and it keeps the journal in local files and SQLite. **memex-lab/memex** — Open-source, local-first AI journal app for iOS and Android. Capture text, photos, and voice — AI agents organize them into timeline cards and insights. Your data stays on your device. Bring your own LLM (OpenAI, Claude, Gemini, Ollama, and more). - Repository: https://github.com/memex-lab/memex - Website: https://memexlab.ai - Stars: 761 · Forks: 76 - Language: Dart - License: GPL-3.0 - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/memex-lab-memex ## The problem Memex targets: journaling fails at capture, not at writing Most journaling apps assume you will sit down and compose. Memex assumes you will not. Its own framing is blunt: it is "Not for writing daily entries, for capturing life in fragments and letting AI organize them." That is the whole design premise, and it decides everything else about the app. The intended user is someone who already scatters notes across a notes app, a camera roll and a voice memo folder, and who wants those fragments turned into something queryable without uploading them to a hosted service. The audience is narrow but real: people who are comfortable pasting an API key into a mobile app and who care that the resulting archive is plain text rather than a row in someone else's database. It is not a Notion replacement, and it is not a diary you hand to a family member. The card types it generates (task, routine, event, person, place, metric, rating, transaction) read like a personal operational log, not prose. ## How the agent orchestration and card pipeline actually work The README describes one conversational agent that coordinates specialist skills and child agents. Those specialists are named explicitly: timeline cards, PKM filing, schedule updates, knowledge insights, diagnostics, comments, memory curation, media handling, and companion chat. So the entry point is a chat surface, and the output of that chat is not just text but writes into the journal's structures. A capture (text, photo, voice) is classified into a card family. Life and productivity records become task, routine, event, duration or progress cards. Knowledge and media records become article, snippet, quote, link or conversation cards. People and places become person or place cards, with map preview. Metrics become metric, rating, transaction or spec sheet cards. Photos land in a gallery. After classification, the README says the system does auto-tagging, entity extraction and cross-reference linking. A second layer sits above the cards. Insight cards are generated to surface connections across records, grouped as charts (trend, bar, radar, bubble, composition, progress ring), narrative (highlight, contrast, summary) and spatial or temporal (map, route, timeline). Knowledge is filed using P.A.R.A: Projects, Areas, Resources, Archives. The storage story is the part worth reading twice. Records, cards and knowledge stay on the device in a filesystem plus SQLite layout, and after AI organization the records "settle into a series of interconnected Markdown files." That is the actual data flow: capture, agent classification, card, then Markdown on disk. If you dislike the agent's filing, the Markdown is your escape hatch, and the README promises one-click export of all records as standard Markdown. ## Installing Memex and getting one card out of it Memex ships as a mobile app, not a server you self-host. The README points to the App Store listing (id6760325170) and the Google Play listing (com.memexlab.memex), and the GitHub releases carry Android Early builds tagged by date and commit, for example android-early-20260910-1b18637. There is no documented Docker image or desktop build in the repository, so treat this as an iOS and Android product. Building from source means a Flutter toolchain. The repository is Dart, with pubspec.yaml at the root and separate android/ and ios/ directories. The README does not spell out the build commands, so the honest starting point is the repository's own BUILD.md, which sits at the top level next to CONTRIBUTING.md. ```bash git clone https://github.com/memex-lab/memex.git cd memex flutter pub get flutter run ``` Those four lines are the standard Flutter sequence implied by pubspec.yaml; the README itself does not print them, so verify against BUILD.md before assuming they are sufficient for a device build. Once the app is running, the first real task is wiring a model, because nothing organizes without one. The README's provider table lists the supported API types, including OpenAI Chat Completions and Responses API, Anthropic Claude API, Google Gemini API, AWS Bedrock, and OpenAI-compatible endpoints for Kimi, Aliyun Qwen, Volcengine Doubao, Zhipu GLM and DeepSeek. You supply the key in the app; prompts go from the phone to that provider directly, and the README states there is no Memex server storing your journal. After the provider is set, capture a short text fragment and let the agent run. What you should see is a timeline card of a matching type, with tags and links attached, and a corresponding Markdown file in your chosen storage location. The README lists iCloud Drive, a custom device folder, and app storage as the three storage options, plus one-tap full backup and restore. ## Where Memex is the wrong tool The provider table contains a warning most reviews would skip. Two rows, Google Gemini via OAuth and ChatGPT via OAuth, are marked "Unofficial, use at your own risk." If your reason for choosing Memex is that you do not want to manage API keys, you are being pushed onto an unsupported path, and the README says so plainly. The release channel is the second constraint. The three most recent releases are all named android-early-*, dated 2026-09-07 through 2026-09-09, with the newest carrying the same commit hash as the one before it. The word early is doing real work there. Anyone who needs a versioned, long-support mobile app should read that naming as a signal. Third, the AI organization is the product. If you want a plain journal that stores what you typed and nothing else, the agent layer is overhead: it consumes tokens on every capture, it decides your card type, and its filing is only as good as the model behind it. Memex's value proposition and its failure mode are the same mechanism. Finally, the companion feature is a different kind of commitment. SillyTavern-compatible character import (V2 JSON plus PNG) and persistent memory across comment and chat scenes mean the app maintains persona state over time. If you want a journal and not a character, that subsystem is surface area you are not using. ## How Memex differs from Obsidian and from hosted AI journals The closest comparison in spirit is Obsidian. Both end up with Markdown files on storage you control, and both expect you to assemble your own workflow. The difference in approach is where the intelligence sits. Obsidian is a text editor with a plugin ecosystem, and any automation is something you install and configure. Memex inverts that: the agent is the primary interface, the Markdown is the byproduct, and the card taxonomy is opinionated out of the box. If you enjoy building your own capture pipeline, Obsidian plus plugins gives you more control. If you never build that pipeline, Memex does it for you and bills you in tokens. The comparison against hosted AI journals is simpler. Those store your entries server-side and bundle the model into the subscription. Memex keeps records, cards and knowledge on the device and asks you to bring the model, which the README lists as OpenAI, Claude, Gemini, DeepSeek and others. The trade is operational: you get an archive you can copy with a file manager, and you accept key management, per-token cost, and the fact that your prompts still travel to whichever provider you picked. Local-first here means no Memex server in the path, not that nothing leaves the phone. Memex AI is the middle option. The README describes it as an optional account-based managed model access that proxies requests and does not store the journal. That is a useful fallback when you do not want to hold keys, but it is a service dependency in an otherwise local app, and the README does not describe its pricing or availability. ## Maintenance, releases and what GPL-3.0 means for you The repository is not archived, and the last push was on 2026-09-09. The release cadence visible in the repository is daily Android Early builds across 2026-09-07 to 2026-09-09, each stamped with a date and a short commit hash. That pattern suggests an actively built project, but it also suggests the Android artifact is a moving target rather than a stable channel, and the README does not describe an iOS release cadence at all. Upgrade cost is the practical question. Because journal content lands as Markdown files plus SQLite, backups are portable, and the README advertises one-tap full backup and restore plus one-click export to standard Markdown. That lowers the cost of leaving, which is exactly the point the README makes about zero vendor lock-in. It does not lower the cost of upgrading: the README does not document a schema migration path, so the safe assumption is that you keep your own export before installing a new build. The licence is GPL-3.0, stated in the README badge and in the LICENSE file at the repository root. For an individual installing the app, that is mostly informational. For anyone who wants to fork Memex and ship a modified version, GPL-3.0 carries source-disclosure obligations, and the repository also ships PRIVACY_POLICY.md and PRIVACY_POLICY_CN.md alongside it. This is a description of what the repository states, not legal advice; if you plan to redistribute a modified build, read the licence text yourself. ## Conclusion Adopt Memex if you already pay for an LLM API and want a journal whose storage you can inspect, back up and export as Markdown. Skip it if you want a zero-setup app, because the README's provider table expects you to supply keys or accept an unofficial OAuth path, and if you need a stable release channel, since the published builds are labelled Android Early. Before committing, install the release, confirm your chosen provider appears in the model list, and export one test entry to check the Markdown output is what you expect. ## FAQ ### Is Memex free? The app itself is open source under GPL-3.0, so there is no licence fee. The AI features are not free in practice, because you bring your own LLM provider and pay that provider, or you use the optional Memex AI managed service, whose terms the README does not describe. ### What is Memex AI used for? It captures text, photos and voice as fragments, then a multi-agent system organizes them into timeline cards, files them using P.A.R.A, and generates insight cards such as charts, narratives and timelines. The README frames it as a capture tool rather than a place to write daily entries. ### Is Memex legit? It is a published open-source project under GPL-3.0 with App Store and Google Play listings, an official site at memexlab.ai, and dated Android Early releases. The README is also explicit that the two OAuth provider options are unofficial and used at your own risk. ### What does Memex do with my journal data? The README states that records, cards and knowledge stay on your device in a filesystem plus SQLite layout, with no Memex server storing the journal, and that prompts go from your phone directly to the LLM provider you configure. The optional Memex AI service only proxies model requests. ### What is Memex used for? The README describes it as an open-source, local-first AI journal for iOS and Android that captures life in fragments and lets AI organize them. Cards cover tasks, routines, events, people, places, metrics, ratings, transactions and photo galleries, with insight cards built on top. ## Sources - [License: GPL-3.0](https://github.com/memex-lab/memex/blob/main/LICENSE) - [memex-lab/memex on GitHub](https://github.com/memex-lab/memex) - [Project website](https://memexlab.ai) - [README](https://github.com/memex-lab/memex/blob/main/README.md) - [Releases](https://github.com/memex-lab/memex/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/memex-lab-memex