Tiered by what you'll
actually capture.

No catch. The daemon is MIT-licensed; the SQLite store is yours; FTS5 search runs locally; the MCP bridge exposes the full read API. Pro adds managed cloud sync (so your engrams follow you to a new machine without setting up an R2 bucket) and the nucleus_ask RAG recall path. If you're willing to run your own R2 bucket and don't need nucleus_ask, Free is genuinely forever.

The daemon keeps running and writing to your local store — that doesn't change. What stops: managed cloud sync (we stop syncing your DB to our R2 bucket) and nucleus_ask (the MCP tool starts returning a "not subscribed" error).

Before you cancel, run eideticd --backups to download your last cloud snapshot — that's a clean local copy of everything that was synced.

Because it's not a quota. The daemon will happily capture a million engrams on the free tier; we don't enforce a cap. The number reflects what most users at each tier actually hit, based on dogfood data. A solo dev using only Claude Code sits around 10K engrams; the same dev across Cursor + Cowork + Claude Code lands around 250K; a team of five pooling context lands at ~250K × 5.

If you blow past the "typical" number, nothing breaks. You don't get throttled, billed differently, or asked to upgrade.

The capture path is live (v0.0.39+: X-Team-ID header dual-writes to engrams/team/<team_id>/<device_id>/ in the managed R2 bucket). The recall path — your AI tool asking "what did the team work on this week?" and getting answers from another seat's engrams — is opt-in per seat and ships behind a feature flag while we work through the cross-seat permission model.

Team-tier today: pooled storage and the API surface to write to it. Cross-seat recall in MCP clients: rolling out across May/June.

Yes, mid-cycle. Pro → Team is prorated on the next charge; Team → Pro takes effect at the end of the current billing cycle (so you keep the seats you've paid for). Cancel any time, no exit form, no "are you sure" dialog.

If you sign up while you're inside the founding window (first 50 Pro / first 10 Team), your price stays at $29/mo or $99/mo for as long as you stay subscribed. If Pro goes to $39/mo next year, you keep paying $29. The lock breaks if you cancel and re-subscribe later — at that point you pay the current price.

It means the daemon's local SQLite store is the source of truth, and nothing leaves your machine unless you turn on sync. On Pro, sync is opt-in (you drop a sync.json; we deliver the API key to enable it). On Free, sync is off entirely unless you bring your own R2 bucket.

The nucleus_ask MCP tool is local-first too — keyword extraction + FTS5 retrieval all happens in the daemon. The host LLM (Claude / GPT / etc.) receives the retrieved engram payloads as context; that's the only point at which engram content reaches a model.

Zero egress fees, S3-compatible, generous free tier, and the daemon's sync code uses a single boring HTTP PUT. BYOR2 means you can wire up your own R2 bucket for free; Pro means we host one for you. If you'd rather sync to S3 / B2 / GCS, that's a config swap; we just don't manage the bucket on those providers.

30 days, full refund, no forms — reply to any email or ping us on Discord and we issue the refund via Stripe (5–10 business days back to your card). After 30 days it's prorated to the day you cancel.

Honest list: (1) custom on-prem deploys with your own ops team — talk to us, that's an enterprise conversation. (2) Per-engram billing or usage-based pricing — we don't do this and don't plan to. (3) Vendor model-routing or "smart" tier selection across LLMs — that's a different product. (4) Cross-tenant analytics — we don't see your engrams.