FAQ
Short answers to the questions people ask most. For how-to detail, see the guides; for the why, see concepts.
Licensing and commercial use
Section titled “Licensing and commercial use”Is OMem free? What about using it at work?
Section titled “Is OMem free? What about using it at work?”OMem is under the PolyForm Noncommercial License 1.0.0. It’s free for personal, academic, research, and non-profit use. Commercial use requires a license — and per PolyForm, “commercial” explicitly includes using it for your for-profit employer’s internal work. If you want to use OMem in a company context, reach out — that’s a conversation worth having.
Why is OMem closed-source?
Section titled “Why is OMem closed-source?”The retrieval, parsing, and curation pipeline is a large body of work, and v1.0 ships as a source-available, closed binary rather than open source. The wiki it produces is fully open — plain Markdown on your disk that you own and can read, edit, grep, and version-control — so you’re never locked in even though the engine isn’t open.
Platforms
Section titled “Platforms”Does it run on Windows or Linux?
Section titled “Does it run on Windows or Linux?”Not in v1.0 — OMem is available on macOS today. It depends on macOS-specific pieces (launchd scheduling, the local Mail/Calendar stores). Windows is coming: the cross-platform groundwork is researched and the core pipeline is already pure Python; it’s a v1.5+ target. Linux follows the same path.
Can it run fully offline?
Section titled “Can it run fully offline?”Retrieval is 100% local — search, page reads, and parsing never touch the network. The one network call is LLM curation at ingest time, which goes to whichever provider you configured. You can point that at a local model (via the openai-compat provider), which works today; the caveat is that curation quality depends on the model. Fully-offline, work-grade quality is a v1.5 goal.
Privacy & data
Section titled “Privacy & data”Is my data really all local?
Section titled “Is my data really all local?”Yes. Your wiki is plain files in a folder you choose; the index and archive are local SQLite + files. Nothing is uploaded, no telemetry is collected. The only thing that leaves your machine is the curator prompt + document text sent to your chosen LLM at ingest — and if that’s a local model, even that stays on-device.
What happens to my data if OMem goes away?
Section titled “What happens to my data if OMem goes away?”Nothing is trapped. Your wiki is a folder of Markdown files you can keep using, reading, and searching with any tool — Finder, grep, Obsidian, git — with or without OMem. The format is open by design (design principle P3).
Setup & dependencies
Section titled “Setup & dependencies”How do I grant macOS permissions? Mail (or calendar) isn’t working.
Section titled “How do I grant macOS permissions? Mail (or calendar) isn’t working.”Mail needs Full Disk Access, which macOS never grants automatically — and on macOS 26+ so does calendar (its database moved into a sandboxed container, so it’s now gated like mail). Run omem setup --grant-tcc — it detects your version, opens System Settings, and highlights the omem binary to add. See common errors → permissions.
Do I need qmd? Should I install it?
Section titled “Do I need qmd? Should I install it?”Strictly required, no — but we recommend it. The built-in fts5 index works out of the box and is fast, but it’s keyword-only: it finds pages that share your literal words (with first-class Chinese segmentation). qmd is where search really comes alive — it adds multi-path retrieval (BM25 + vector embeddings + query expansion + reranking), so it matches by meaning, handles cross-language queries, and is markedly more accurate on real, messy work. The cost is one-time: a separate binary and ~2.2 GB of local models on first use, all running on your machine. If you’re deciding, install qmd — fts5 is the dependable fallback, not the better experience. See retrieval for how it works.
What does ingest actually cost in LLM calls?
Section titled “What does ingest actually cost in LLM calls?”Curation calls the LLM once per new or changed item — and the curation cache means unchanged items cost nothing on re-runs. So total cost tracks “how much new content,” not “how many times you ran ingest.” A steady mailbox costs little after the initial pass; the expensive case is forcing a full re-scan with --now.
How it compares
Section titled “How it compares”How is this different from mem0 / Letta / Zep?
Section titled “How is this different from mem0 / Letta / Zep?”Those are chat-memory SDKs — they remember what was said to an agent. OMem reads your actual work: emails, documents, calendar, meeting notes. Different problem. See what OMem is not.
How is this different from Microsoft 365 Copilot / Glean?
Section titled “How is this different from Microsoft 365 Copilot / Glean?”Those are enterprise products requiring IT deployment and admin consent, and their index only feeds their assistant. OMem is local-first (no IT ticket), and its memory feeds any agent you use — Claude Code today, whatever you switch to next. Copilot can reach your M365 data but tops out at “assist”; OMem hands context to agents that can actually do the work.
How is this different from Obsidian / Notion?
Section titled “How is this different from Obsidian / Notion?”Those store what you manually put in them. OMem auto-ingests the work already scattered across your machine — the inbox you live in, the synced OneDrive folder — without you re-typing anything.
How is this different from OpenViking?
Section titled “How is this different from OpenViking?”OpenViking (ByteDance) is the closest neighbor on the technical axes — it is local-first and agent-agnostic, and OMem even borrowed its tiered context-loading idea. The difference is decisive: you feed content into OpenViking programmatically, whereas OMem auto-ingests the real work already on your machine — your inbox, your synced folders, your PDFs. See what OMem is not.
How is this different from GBrain?
Section titled “How is this different from GBrain?”GBrain shares the philosophy — markdown-as-memory, agent-agnostic — but it’s bring-your-own-markdown: you write the notes. OMem builds the wiki for you, from your actual office work.
Roadmap
Section titled “Roadmap”When is <feature> coming?
Section titled “When is <feature> coming?”The two headline efforts after v1.0:
- A graphical interface — v1.0 is CLI-first; a proper desktop UI (so you don’t live in the terminal) is the biggest piece of work ahead.
- An entity wiki — Layer 2 — today every page is rooted in one source. The schema already reserves a second layer: pages distilled across sources into one per person, project, or topic, with graph-style relationships linking them back to the evidence. This is a flagship direction, not a footnote — the groundwork (
layer/source_backlinkscolumns) ships in v1.0.
Also on the way:
- Windows / Linux — the cross-platform groundwork is researched and the core pipeline is already pure Python.
- More mail sources — Outlook Classic / Outlook on the Web / AppleScript, beyond today’s Mail.app.
- More sources — Slack, Jira, and other tools where work lives.
- Cross-device sync — on the list, not in v1.0.
Still stuck?
Section titled “Still stuck?”Run omem doctor first — it catches most problems and prints a fix. Then check common errors. For anything else, the GitHub issue tracker is the place.