Frequently asked questions

Memrith is the continuity layer for AI — a desktop app that builds a long-term, editable model of your work and thought, and lets the AI you use draw on it across months and years instead of one conversation at a time. Every entry you write and document you upload is read by your AI provider and folded into a structured personal database, so when you talk to the AI it knows the arc of what you've been doing, not just the prompt in front of it.

Your entries, memory, and conversations live on your machine. The only place they go is the AI provider you choose (Anthropic, OpenAI, or OpenRouter), as the messages you send — direct from your device on your own key. Memrith also makes routine license and app-update checks, but those carry nothing about your data; see how your data flows for the full list.

vs ChatGPT / Claude: ChatGPT and Claude now carry some memory across chats, but it lives in their cloud, isn't a file you can fully open and edit, and can't move to another provider. Memrith keeps a structured memory of everything you've told it — goals, facts, patterns, projects — on your own machine, and feeds the relevant pieces back to the model on every conversation. The memory is editable and yours: you can correct what it learned and watch it use the corrected version going forward, and it carries with you when you switch models.

vs Notion / Obsidian: Those are documents you write and read. Memrith is a personal database the AI maintains for you. You can still browse and edit it, but you don't have to organize it — the AI extracts the structure from raw input.

People who want an AI assistant that remembers context across days, weeks, and months — researchers, writers, founders, students, anyone doing long-running creative or analytical work. If you're tired of re-explaining your project every time you open a chat, that's the gap Memrith fills.

It's not a fit for: quick one-shot Q&A (just use the chat directly), team collaboration (Memrith is single-user / single-device), or anyone who doesn't want to manage their own AI provider account.

The Memrith app source is closed (it's commercial software). Some bundled components are open source under their original licenses; the full third-party attributions are available on request at support@memrith.com.

You don't need the source to verify the privacy story: Memrith is local-first and BYOK by design — see Architecture and Why BYOK for exactly what data leaves your machine (and what doesn't).

$9.99 — one-time. No subscription, no monthly fee, no per-message charge from Memrith. You pay once, the app is yours.

AI usage is billed separately by whichever provider's API key you supply (Anthropic, OpenAI, or OpenRouter) — typically a small amount per day for normal personal use. Set a hard spending cap in your provider's dashboard if you want a guarantee.

We don't offer a separate free trial, but Memrith ships with a 60-day money-back guarantee — full refund within 60 days of purchase, no questions asked. That's effectively a 60-day trial with the same risk profile (you pay, then decide).

See the refund policy for full terms.

Email support@memrith.com within 60 days of purchase and we'll refund you in full. No quiz, no friction. Polar (our payment processor) refunds back to your original payment method.

Processing takes about 5 business days on our end; your bank may add another 3–10 business days. Full terms in the refund policy.

No. Memrith charges the one-time $9.99 and nothing else — no per-call fees, no premium tier, no future "pro" upgrade pricing.

AI usage costs go directly to your AI provider (Anthropic / OpenAI / OpenRouter), billed by them based on the tokens you use. We never touch that money or take a cut.

No. All updates to Memrith — bug fixes, new features, performance improvements — are included for the life of your license. The auto-updater fetches them when available; you don't have to do anything.

A single license activates on up to 2 machines (personal-use). If you need more, email us and we'll work something out.

Two steps:

1. On the old machine: open Memrith → Settings → License & Purchase → Deactivate license. This frees the activation slot.
2. On the new machine: install Memrith → Settings → License & Purchase → Activate license → paste the same key.

If you no longer have access to the old machine, email support@memrith.com with your order ID and we can free the slot manually.

Your license key was emailed to you at the time of purchase by Polar (subject: "Your Memrith license"). Check your email for it — including the Promotions / Spam folder.

If you genuinely can't find it, go to polar.sh/purchases and log in with the email address you used at checkout. Your past orders + license keys are listed there.

No. Once you've activated, the license is valid indefinitely.

(Memrith periodically re-checks the license with Polar in the background to detect cancellations / refunds. If you initiate a refund, the license deactivates automatically.)

No. Your entries, your memory database, and your continuity all live on your machine. Switching providers swaps which model reads your memory — nothing about the memory itself changes.

The flow: Settings → AI Provider → pick a different provider → paste its key → save. Your accumulated context is intact. The next chat opens with the same memory, just routed through a different model.

This is the whole point of the BYOK architecture. Why Memrith is BYOK by design →

You change providers in Settings. Your memory is untouched. This is the entire reason Memrith is BYOK by design — your continuity layer is independent of any single AI company.

Translation: the provider becomes replaceable infrastructure. The memory you've built up over months or years doesn't depend on any one vendor staying in business, keeping their pricing, or maintaining your account.

Yes — since v1.3.0. Run a local AI server — Ollama, llama.cpp, or LM Studio — and point Memrith at it in Settings → AI Provider → Local, or during first-run setup. No API key is needed for most local servers, and inference is free.

One honest note: this is bring-your-own-server. You install and run Ollama (or its equivalents) yourself; Memrith connects to it. In return, Memrith is built for the smaller models you run locally — it sizes each request to your model's context window (you can set the size), tidies minor formatting slips in responses, and allows the longer response times local models need. Switch back to Anthropic, OpenAI, or OpenRouter any time — your memory comes with you either way.

Three reasons:

1. Cost transparency. AI calls are paid directly to the provider, billed on actual usage. There's no Memrith markup, no monthly minimum, and you can set hard caps in the provider's dashboard.
2. Privacy. Memrith never sees your conversations — they go directly from your machine to the AI provider you chose. There's no Memrith server in the middle that could be subpoenaed, breached, or change its terms.
3. Portability. Your continuity stays yours even if a provider raises prices, restricts access, gets acquired, or shuts down. See the model-switching question above.

Trade-off: you have to do a 2-minute setup with an AI provider before Memrith is useful. See the walkthrough.

Memrith works with three: Anthropic (Claude), OpenAI (GPT-4o / GPT-4o-mini), and OpenRouter (one key, route to many models). All three are supported equally — we don't get a kickback from any of them, so pick whichever fits.

Some considerations:

• Anthropic — Claude models. Pay-as-you-go.
• OpenAI — GPT models. Pay-as-you-go.
• OpenRouter — Aggregator; one account/key gives you access to many models from many providers. Slightly more flexible if you want to experiment.

Pricing varies between providers and models — check their pricing pages before signing up. You can switch providers any time in Settings → AI Provider without losing any of your Memrith data.

Depends on the provider, model, and how much you use Memrith. For normal personal journaling and chat use, the typical range is single-digit dollars per month.

What drives cost up:

• Choosing a high-end model (Claude Opus, GPT-4o) instead of a fast/cheap one (Claude Haiku, GPT-4o-mini)
• Heavy memory work (very long uploads, frequent re-syntheses)
• Long chat sessions with lots of context

Set a spending cap in your provider's dashboard. Anthropic, OpenAI, and OpenRouter all support hard limits — if you cap at $10/month, you cannot be charged more than $10/month. This is the only guaranteed way to control cost.

No, unfortunately. ChatGPT Plus and Claude Pro are web app subscriptions and don't include API access. The AI API is a separate paid product from the same companies; you sign up at console.anthropic.com or platform.openai.com and pay-as-you-go.

The good news: API usage is usually much cheaper than the $20/month web subscriptions for normal personal use — you only pay for what you actually call.

Yes. API keys are stored in your operating system's keychain (macOS Keychain Access, Windows Credential Manager, libsecret/KWallet on Linux) — the same secure store your browser uses for saved passwords. Memrith never logs, prints, or transmits the key anywhere except direct API calls to the provider.

If your OS doesn't have a usable keychain backend, Memrith falls back to an owner-only file (chmod 0600) in your local data dir with a logged warning. Either way the key stays on your machine.

• macOS 11 (Big Sur) or later — Apple Silicon (M1 or newer). Intel Macs are not supported, and Intel support is not currently planned.
• Windows 10 or later (64-bit)
• No Python install required — Memrith bundles a private Python runtime; there's nothing to set up.
• ~500 MB free disk space
• Internet connection — only for direct calls from your machine to your chosen AI provider. No Memrith server, no Memrith account.

This is normal for apps downloaded outside the Mac App Store. It is not a sign anything is wrong with Memrith. To clear the prompt the first time:

1. Right-click (or control-click) Memrith.app in your Applications folder
2. Choose Open from the menu
3. A dialog appears asking if you're sure — click Open

You only need to do this once. Every future launch works with a normal double-click.

(Memrith isn't currently Apple-notarized — that's why the prompt appears. The updates the app installs are still cryptographically signed and verified before they run.)

Windows SmartScreen flags any executable Microsoft hasn't seen before — it's a "is this a popular app?" check, not a malware scan. As more people install Memrith the warning goes away automatically. To clear it the first time:

1. Double-click Memrith-Setup.exe
2. If a blue "Windows protected your PC" dialog appears, click the small More info link
3. A Run anyway button appears — click it
4. If a separate User Account Control prompt asks permission, click Yes

One time per machine.

Memrith's core logic — memory extraction, search, the AI gateway — is written in Python, a free, open-source language maintained by the non-profit Python Software Foundation.

You don't need to install it. Memrith bundles its own private Python runtime inside the app, so there's nothing to set up, no PATH to configure, and no separate download.

Most common causes:

• Antivirus quarantined part of the install. Check your antivirus quarantine for items in the Memrith folder, restore them (or add an exception), and relaunch.
• WebView2 runtime missing (Windows 10). If the window won't open, install or repair Microsoft's free WebView2 runtime, then reopen Memrith.
• Incomplete copy or download. Reinstall from the link in your purchase email — the runtime and all dependencies are bundled, so a fresh install fixes a damaged one.

If none of those, Memrith writes a diagnostics file and a log you can send us — on macOS in ~/Library/Application Support/Memrith/Logs/, on Windows in %APPDATA%\Memrith\Logs\. Email diagnostics.txt (or the last 50 lines of Memrith.log) to support@memrith.com and we'll diagnose.

All your entries, memories, and uploads live in:

• macOS: ~/Library/Application Support/Memrith/
• Windows: %APPDATA%\Memrith\
• Linux: ~/.local/share/Memrith/

It's all plain JSON files — readable in any text editor. Back this folder up to preserve your memory; restore by copying it back.

No. There is no Memrith server. Your entries are written to disk on your device, processed locally, and sent only to the AI provider (Anthropic / OpenAI / OpenRouter) whose API key you supplied — through a direct connection from your machine to them.

The one optional exception is crash reporting, which is off by default and you have to explicitly opt in. Even when enabled, it scrubs API keys, file paths, and journal content before sending. Full details in the privacy policy.

Yes — Memrith works by sending relevant entries / memories to the AI provider as context for chat and extraction. That's unavoidable: the AI needs the content to do anything useful with it. What gets sent is governed by that provider's own privacy / data-use policies:

• Anthropic Privacy Policy
• OpenAI API Data Usage Policy (note: API data is NOT used to train OpenAI models by default)
• OpenRouter Privacy Policy

None of the three use API-submitted data for training by default. You can confirm + tighten settings in each provider's dashboard.

Memrith creates an automatic snapshot of each workspace before any major operation (smart updates, migrations). Up to 10 regular + 5 migration backups are kept per workspace in the workspace's backups/ folder.

For full backups: Settings → Storage → Create backup writes a timestamped snapshot you can restore from at any time. To move data between machines, Settings → Storage → Export workspace writes a .zip of the entire workspace — drop it into Import on another machine.

Quit Memrith, then delete the Application Support folder shown in "Where is my data stored?" above. Next launch will run a fresh first-time setup.

This is irreversible. If you want to start fresh but keep the option to recover, use Settings → Storage → Export workspace first to save a .zip backup somewhere safe.

Memrith's JSON files on disk are not encrypted by Memrith itself — they rely on your OS's standard disk encryption (FileVault on macOS, BitLocker on Windows) for at-rest protection. Turn those on if you haven't.

API keys are stored encrypted, in the OS keychain.

Three steps, repeated continuously:

1. You write a journal entry or upload a document. Memrith sends it to your AI provider.
2. The AI extracts structured "memories" — facts, goals, patterns, key claims, named entities — into your personal database. These are stored as editable JSON records, each linked back to the source they came from.
3. When you chat or ask a question, Memrith pulls relevant memories into the context window so the AI has the background it needs to answer well.

The memory grows over time. You can browse, search, and edit it directly in the Memory and Database tabs.

Open the Memory or Database view, find the memory, and edit or delete it. The AI uses the corrected version on the next request — your edit is the source of truth.

If something persistently shows up wrong, edit the original entry that produced it (in the Entries view) and trigger a re-extraction. Otherwise the AI may re-derive the same wrong claim from the original source.

Yes. Memrith accepts:

• .txt — plain text (chat exports, notes, anything)
• .pdf — text-extracted automatically
• .docx — Microsoft Word documents
• Pasted text directly into the upload box

Large files are chunked and processed in parallel — very long documents work fine. Find the uploader under Settings → Upload documents, or during the first-launch onboarding.

Partly. You can browse, search, and edit existing memory + entries without internet. But anything that requires the AI (new chat, processing a new entry, extracting memories from an upload) needs to reach your AI provider's API — so an internet connection is required for those operations.

If you enable Local Semantic Search in Settings, the embedding model runs entirely on your device with no API calls. That doesn't replace the chat, but it does make memory search work offline.

Yes. Settings → Workspaces lets you create separate workspaces — e.g. one for personal journaling, one for a specific project — each with its own memory database, uploads, and entries. Switch between them from the same Settings panel; data is fully isolated.

You haven't connected an AI provider yet. Go to Settings → AI Provider and paste a key for whichever provider you want to use. See the step-by-step API-key walkthrough if you don't have one yet.

Two common causes:

• Whitespace in the key. When you copied the key from the email, you may have grabbed a leading or trailing space. Re-copy carefully — the key is exactly 36 characters, UUID format (xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx).
• Wrong product key. If you bought a different product from the same Polar account, double-check which key you're pasting.

If neither of those, email support@memrith.com with your order ID — we can verify in the Polar dashboard.

Almost always an AI provider issue, not Memrith. Try:

• Check the provider's status page. Anthropic / OpenAI / OpenRouter. Outages there cascade into Memrith.
• Switch to a faster model in Settings → AI Provider. Claude Haiku / GPT-4o-mini are much faster than their flagship models for everyday work.
• Check your network. AI API calls are blocked on some corporate / school networks; try a different network as a quick diagnostic.

• macOS: ~/Library/Logs/Memrith.log
• Windows: %LOCALAPPDATA%\Memrith\Logs\Memrith.log
• Linux: ~/.local/share/Memrith/Logs/Memrith.log

If you're reporting a bug, attach the last ~50 lines (the part right around when the issue happened). Memrith never logs API keys, full file paths, or journal content — the log is safe to share.

Quit Memrith. Delete the Application Support folder (see "Where is my data stored?"). Next launch runs first-time setup again.

This deletes everything — all entries, memories, configuration, the lot. If you might want any of it back, export a workspace zip first.

Memrith checks for updates in the background once per day. If you want to force a check, you can manually download the latest from the releases page and re-install — your data + license stay intact.

If the auto-updater shows an error, the most likely cause is a corrupted or invalidated release manifest signature; an investigative reinstall from the releases page always resolves it.