# Security and your data

Where your memory is stored and processed, who can reach it, how we test the service and how to report a problem. This page says only what our code and our legal pages already establish, names its source for each part, and dates each check run on the live host.

## In short

Stored: on Google Cloud in the us-central1 region (Iowa, United States). Not in your browser and never in a repository.

Processed: to search by meaning, the text of each note when it is indexed, and each search query, go to Google Vertex AI in the same region, which returns embeddings (lists of numbers that describe the meaning). To name the topics of the dashboard’s memory map, a short sample of each topic’s memories goes to Gemini on Google Vertex AI, in its United States multi-region, which returns the names. As checked on 23 October, no hosted workspace sends memory to any other AI provider; the AI clients you connect receive what they ask for.

Yours to take or remove: export everything as Markdown at any time, delete retired memories yourself on the dashboard, and delete your account by email.

Fully private: run the open-source engine on your own computer. With no model key set, nothing leaves your machine.

## Where your memory lives

Your memory content is stored on Renor-controlled infrastructure hosted on Google Cloud, currently in the us-central1 region (United States), and encrypted at rest. The dashboard shows it in your browser when you ask, but it is not stored there, and it is never in a public repository. Account records (sign-in, workspace, plan) are in Supabase; the site and dashboard are served by Cloudflare Pages.

Each workspace has its own memory directory, its own search index and its own server process. That process listens on the host’s loopback interface only, behind the one proxy that terminates TLS and the gateway behind it, so it is not reachable from the internet directly.

Source: Privacy policy §3 and §7; Data protection §4; engine/apps/gateway/README.md; cloudflare/README.md.

## Who else sees your memory text

Only the processors that are needed to run search and to show your memory in the dashboard. Supabase's function runtime relays memory text and queries between the dashboard and the memory service (saving, recalling, browsing, reading and exporting; an import goes straight to the memory service) and does not store them. Every hosted workspace has search by meaning on, so the text of notes and the queries needed for it go to Google Vertex AI (gemini-embedding-001, 768 dimensions), in the us-central1 region, and a short sample of each topic’s memories goes to Gemini on Google Vertex AI, in its United States multi-region, to name the topics of the dashboard’s memory map. Google’s Vertex AI terms say Google does not use this data to train or fine-tune its models without the customer’s permission, and Renor has not given it. Write-time summaries, which would send each saved note to Fireworks AI (US-only endpoint), are switched off by default and used only where a deployment turns them on. No workspace has an evidence-extraction model configured.

Analytics events describe actions, not content: no memory text, prompt or query is sent to analytics or to error reporting. Payments are handled by Paddle as merchant of record; card details never reach Renor.

We do not use your memory content to train models unless you opt in to the data level, we do not sell it, and we do not share it with other users.

Sharing with Renor: anonymous usage numbers are on by default, on every self-hosted install and hosted workspace, to improve Renor for everyone. A self-hosted install tells you in a one-line notice before anything is sent: in the first reply it gives your agent, and in the log of a service you run in the background. A hosted workspace shows no separate notice; this page is how it tells you. The default level, signals, sends only anonymous labelled numbers: an install record (a random install id, the engine version and commit, the Node major version, the operating-system family and architecture, whether it runs in CI, and how it is connected); a daily heartbeat (the AI client’s name and version, which embedding, reranker and index are in use, the memory store’s size in buckets, and per-tool call counts, latency buckets and sanitized error classes); the steps of getting started, each with its hour, under a random id made for each run; a count of opt-outs that carries no id; and numbers worked out inside your own data folder from the learning log: retrieval results and later outcomes, save outcomes, and session totals and timing. Every field is listed under “Every field sent” in the engine’s telemetry page, at github.com/renorai/renor-engine/blob/main/docs/TELEMETRY.md. When the numbers arrive, Renor works out the two-letter country code from the connection and stores only that code. We never store your IP address with your data. To stop abuse, a scrambled one-way code of it is kept for at most 2 days, separately, and then deleted. It sends no memory text, queries, memory ids, keys, names, file names or paths. To turn the numbers off on a self-hosted install, set RENOR_SHARING=off or DO_NOT_TRACK=1, run renor-engine telemetry off, or tell your agent; for a hosted workspace, write to privacy@renor.cloud. The data level sends those signals plus the query behind a retrieval, the text of the memories it returned, and the statements you saved. It is never on by default: text is shared only after you explicitly say yes, at a terminal on a self-hosted install or with the “Improve Renor for everyone” switch in a hosted workspace’s dashboard. Renor may use those examples to train, evaluate and improve its own models. Shared data is kept for as long as it helps improve Renor, with no fixed period, and is deleted when you ask or when you turn the text switch off.

Source: Privacy policy §2–§4; supabase/functions/memory-inventory/index.ts; src/services/import.ts; engine/scripts/alpha/enable-tenant-embeddings.sh; Google Cloud’s Vertex AI data-governance page (docs.cloud.google.com/vertex-ai/generative-ai/docs/data-governance).

## Running Renor yourself

The memory engine is open source, at github.com/renorai/renor-engine, and runs on your own computer with Node 22. Your memory is then a folder of Markdown files that you can back up, sync or move like any other folder, and no Renor server is involved.

Nothing leaves your machine unless you set a model key. With none set, search matches exact words, and saved facts are found either way. If you set an OpenAI key for search by meaning, your notes’ text goes to OpenAI to be embedded; the engine’s README lists every setting that sends anything out. The hosted service adds the website, dashboard, sign-up and operations around the same engine.

Source: the renor-engine README (“Search without embeddings”, and what leaves your machine); How it works, “What Renor does not do today”.

## Keeping workspaces apart

The gateway looks up which workspace a connection belongs to on every request, and refuses to route when it cannot (a missing or unknown workspace fails, it never falls back to another one). Revoking a connection token takes effect on the next request.

Each workspace is kept separate, and the self-serve provisioner sets up new ones the same way: its files cannot be read by another workspace’s process. One limit remains: the gateway and the workspace engines still share one operator credential (the gateway secret), so isolation between workspaces rests partly on the gateway and engines refusing requests; per-workspace credentials are not in place yet. The engine and the gateway fail closed: a request without a valid owner token reads and writes nothing.

Source: engine/scripts/alpha/README.md (“Per-tenant service accounts”); engine/apps/gateway/README.md; engine/docs/architecture/MCP-TOOL-SURFACES.md.

## Signing in and connecting

An AI client connects with the standard MCP sign-in (OAuth 2.1 with PKCE) as the workspace owner. Connection tokens are credentials: they are stored hashed and are revocable from the dashboard, which stops that client at once.

Source: engine/src/mvp/auth.ts; Privacy policy §2 and §5.

## Export and portability

export_memory returns everything the workspace holds as portable Markdown, in one documented format (renor-export/v1): the fact ledger exactly as stored, every daily-log entry, every note in full, and everything retired, each with its time, scope and source where the engine recorded them. A manifest counts each kind and says where to continue; a large workspace is paged, never cut. Not exported: the search index and the embeddings, which are derived from the Markdown and rebuilt from it. See How it works, “Export and import”.

Source: engine/src/mvp/exportAll.ts (export_memory); engine/src/memory/lifecycle.ts (ledger events); engine/src/memorymap/memoryMap.ts (daily entries, retired items); engine/docs/architecture/EXPORT-FORMAT.md.

## Retention, backups and deletion

We keep your data while your account is active. Retiring a memory removes it from every read and search and can be undone; it is not erasure. You can then delete a retired note, log entry or fact permanently on the dashboard’s Memories page, and that cannot be undone; erasing a fact leaves one line in your workspace’s history record with none of its words (the event id, the time and a digest).

You can delete your account yourself under Settings → Your data → Delete account in the dashboard: you type your workspace’s name to confirm, and we cancel any paid plan, revoke every connection key and agent sign-in, erase your workspace and every memory in it with its history from our server, delete the account and email you when it is done, retrying any step that fails. Where Settings does not offer the button yet, or if you prefer, email privacy@renor.cloud from your account address and we do it by hand and confirm. Our analytics, error-reporting and email providers keep their own logs for their own retention periods; deleting your account does not erase events already sent to them.

Paddle, our merchant of record, keeps its own payment records as the law requires; deleting an account cancels the subscription but does not erase those records.

A snapshot policy snapshots that disk daily at 08:00 UTC and keeps each snapshot for 7 days. We have not yet validated deletion against restored snapshots, so we do not claim that deleted data is gone from backup copies.

You can change or withdraw sharing at any time. For a hosted workspace, turn off “Improve Renor for everyone” in the dashboard to stop sharing memory text, or write to privacy@renor.cloud to stop the usage numbers as well; for a self-hosted install, run renor-engine telemetry off: sharing stops at once and Renor is asked to delete everything it holds for that workspace or install. A self-hosted user can instead run renor-engine telemetry delete to ask for deletion without changing the level; anyone can contact privacy@renor.cloud. If the endpoint cannot be reached, the request is kept and retried, and backups can hold a copy for up to 7 more days. A ranking setting already fitted from shared rows cannot be recalled, but the next fit leaves your rows out.

Source: Privacy policy §5; Data protection §8; docs/DATA-DELETE-RUNBOOK.md.

## Independent review

Renor has not yet completed an independent security audit. We plan to commission one as the service grows. Until then, we test it ourselves on a schedule, described below, and welcome reports at security@renor.cloud.

## How we test our own security

Since early October 2026 we have run our own adversarial reviews: AI agents attack a local copy of the service and the public site, one area at a time — keeping workspaces apart, sign-in and OAuth, prompt injection through stored memory, quotas and metering, and what leaves the system. Each confirmed issue is fixed in order of risk, and the attack is kept as a standing test so it cannot quietly come back.

Source: engine/tests/breaker; scripts/selfserve (attack tests).

## Reporting a security issue

To report a vulnerability or anything else that puts users’ data at risk, email security@renor.cloud. The same address is published at renor.cloud/.well-known/security.txt. We aim to reply within three working days to say we have it and what we will do, and we tell you when it is fixed.

Please tell us before you make an issue public, and do not access or change data in a workspace that is not yours.

Source: public/.well-known/security.txt.

## Availability

Live status and uptime history: https://status.renor.cloud

There is no uptime guarantee (SLA) on the current plans. A call that fails because the workspace is unavailable writes nothing and is not queued (see How it works, “When the workspace is asleep or unreachable”).

If a breach affects your data we will notify you and the authorities as the law requires. Questions about any of this: privacy@renor.cloud.

Source: Terms §7 and §9; Privacy policy §7; status.renor.cloud.

## Live checks, and when they ran

A sentence about live infrastructure can change without a code change, so each one that is verified on the host is recorded here with its date. The date is when the check ran, not an expiry: an older date means an older check, not a withdrawn claim.

2026-10-07: the disk was encrypted at rest; each live workspace was kept separate; every hosted workspace had search by meaning on; and the daily snapshot policy had produced snapshots for October 4 to 7.

2026-10-23, after the 22 October server move: memory was hosted on Google Cloud in us-central1 (United States) and encrypted at rest, with a daily backup kept 7 days in the same region; each workspace was kept separate; every one sent search by meaning to Google Vertex AI in us-central1, and none was configured to send memory to any other model provider.

Source: checks run on the live host. Each line is written on the day its check runs.

See also [how it works](https://renor.cloud/how-it-works), the [Privacy Policy](https://renor.cloud/privacy), [data protection](https://renor.cloud/lgpd) and the [Terms of Service](https://renor.cloud/terms).

[Renor home](https://renor.cloud/) Durable memory for AI. Your agents stop starting from zero.

Product [How it works](https://renor.cloud/how-it-works) [Docs](https://renor.mintlify.app) [Research](https://renor.cloud/research) [Culture](https://renor.cloud/culture) [Press](https://renor.cloud/press) [Install](https://renor.mintlify.app/install) [Blog](https://renor.cloud/blog/) [Pricing](https://renor.cloud/pricing) [Status](https://status.renor.cloud)

Account [Get started](https://app.renor.cloud/signup) [Sign in](https://app.renor.cloud/login)

Legal [Security](https://renor.cloud/security) [Privacy](https://renor.cloud/privacy) [Data protection](https://renor.cloud/lgpd) [Terms](https://renor.cloud/terms) [Refunds](https://renor.cloud/refund) [Support](https://renor.cloud/support)

© 2026 Renor

- [Renor on X](https://x.com/renor_ai)
- [Renor on LinkedIn](https://www.linkedin.com/company/renorai/)
- [Renor on YouTube](https://www.youtube.com/@renor_ai)
- [Renor on Twitch](https://www.twitch.tv/renor_ai)
- [Renor on Instagram](https://www.instagram.com/renor.ai/)
- [Renor on Threads](https://www.threads.com/@renor.ai)
