Plasmole
  • The work Design Plan and simulate Verify Record Analyze
    Around the work AI Run the lab Library and connections
    Also in Product Integrations Chat tools (MCP) All tools
  • By organization
    Academic labs Biotech startups Contract research organizations
  • Security
  • Pricing
  • About
Login Get Started
Product Product overview Design Plan and simulate Verify Record Analyze AI Run the lab Library and connections Integrations Chat tools (MCP) All tools
Solutions Academic labs Biotech startups Contract research organizations
Security Pricing About
Login
Legal

Privacy Policy

Last updated: November 2026

On this page Introduction Information We Collect MCP Connector Data Handling Connected Storage: Google Drive How We Use Your Information Data Security AI Processing Cookies and Analytics Data Sharing Your Rights Data Retention Changes to This Policy Contact Us

Introduction

At Plasmole, we are committed to protecting your privacy and ensuring the security of your personal information. This Privacy Policy explains how we collect, use, and safeguard your data when you use our plasmid design platform.

Information We Collect

Account Information

When you create an account, we collect your name, email address, and password. If you sign up on your own rather than through an invitation to a lab, your first visit asks your role (and your name, if the account has none) and, if you choose to answer, your lab or company, the organisms and cloning methods you work with, the tools you use today, your goal, and how you heard about us. You can change the answers in Settings. We store them with your account. The assistant reads your role, organisms, methods, tools and goal, and on older accounts the vectors listed during an earlier setup, to fit its answers to your work. Our analytics provider receives only the answers you pick from a list, never what you type. If you later create or join a team workspace, we store that membership so the workspace can show who belongs to it.

Research Data

We store the plasmid designs, sequences, and related research data you create or upload to our platform. This data remains private to your account unless you explicitly choose to share it.

Voice Recordings

When voice input is switched on, a microphone button appears in two places: the AI assistant's message box, and the toolbar of a notebook entry you can edit. It records only while you have started a recording with it, and it is not offered in a workspace where AI processing is turned off. When you stop, the recording is sent to our server and forwarded to our transcription provider. In the assistant, the text comes back into your message box for you to read and edit before you send it. In a notebook entry, the text is put into the entry where your cursor was, or at the end of the entry if there was no cursor, and you can edit or undo it like anything you type there.

We do not store the recording. It is not written to our database, our file storage, or our backups, and there is no temporary copy that a cleanup job deletes later: the audio exists only in memory for the length of that one request. What persists is the text. In the assistant, that is only once you send it as a message, at which point it is part of your assistant conversation like anything else you type; if you never send it, nothing is kept. In a notebook entry, the text is part of the entry as soon as it is put in, and is saved with it like anything else you type there.

Our transcription provider is OpenAI, which is already one of the two model providers described under “AI Processing” below. OpenAI transcribes the recording in the United States, wherever you are, and we have no option that keeps it elsewhere. Their handling of the recording is governed by our API agreement with them; we do not train models on it.

Usage Information

We collect information about how you use our platform, including features accessed, time spent, and general usage patterns to improve our services.

MCP Connector Data Handling

If you use Plasmole through a chat AI (Claude, ChatGPT, or any other host that supports the Model Context Protocol Apps standard), our backend processes only what's strictly required for the tool to do its job. We do not log the chat conversation that surrounds the tool call.

Connecting requires you to authorize the connector against your Plasmole account, so connector activity is tied to your account in the same way activity in the web app is. An earlier anonymous tier was retired in July 2026.

What we collect (MCP)

  • Tool arguments and responses. The structured input you (or the LLM on your behalf) sent to a Plasmole tool (for example, the DNA sequence you asked us to annotate) and the response we generated.
  • Source-channel attribution. The host's user-agent string is parsed into a coarse channel label (chatgpt_app, claude, mcp_unknown) and stored with each plasmid we create on your behalf. We also store the raw user-agent string the chat host sent, so we can tell channels apart when a vendor changes it. It identifies the client software, not you.
  • Your Plasmole account identifier. Using the connector requires you to authorize it against your Plasmole account, so every call is attributed to that account. The identifier we retain is your Plasmole account UUID: it is stored on any pending plasmid record your call creates, and the same UUID is sent to our analytics provider so we can measure how many connector sessions go on to save a plasmid. It is the identifier already associated with your account, not a derived pseudonym. Signed-in connector calls are subject to the same hourly request ceiling as the rest of our API, though the counter itself is keyed on the calling IP address rather than on your account (see the IP bullet below). The per-day and per-tool caps described in earlier versions of this policy applied to the anonymous tier retired in July 2026.
  • A per-call usage record. Every connector tool call writes one row to our internal cost and usage ledger. That row records your Plasmole account identifier, the name of the tool you called, the channel label above, the authentication tier the call used, when it started and finished, and the compute and model cost we attribute to it. We use it to meter usage against your plan, to bill, and to understand which tools are worth maintaining. It is a record that you called a given tool at a given time; it does not store the sequence you sent or the response we returned.
  • The calling IP address, in two places. First, telemetry for requests rejected before authentication: calls presenting no bearer token, or one we cannot validate. For those we record a salted, truncated one-way hash of the address rather than the address itself, so we can tell one rejected caller from another, and that hash is not attached to any plasmid record. Second, and this applies to authenticated connector calls as well: our request-rate limiter keys its hourly counter on the caller’s IP address whenever it cannot resolve the request to an account from the bearer token alone. Connector tokens are opaque rather than signed, so in practice this is every connector call. The address is held in that counter in our cache for the duration of the one-hour window and then expires. It is the raw address there, not a hash, and it is not written to any plasmid or ledger record.
  • We describe the hash above as a pseudonym rather than as anonymization: it separates callers, and we do not claim that someone who already holds a candidate address could not check it against the hash.

What we do NOT collect (MCP)

  • Conversation content. We do not see, log, or store the text of your conversation outside of the explicit tool arguments. The LLM's reasoning, system prompts, and the rest of the chat thread never reach our backend.
  • Payment information. The connector does not collect payment details. Billing, where it applies, is handled through your Plasmole account.

Pending plasmids

When you use a tool that produces a plasmid, we create a temporary pending_plasmid record holding the sequence plus annotations, with a bearer token. The token is the URL embedded in the “Open in Plasmole →” button. If you click that button and create a Plasmole account, the plasmid materializes into your library. If you don't, the record expires 24 hours after creation and stops being reachable through the link. We are candid about a gap here: the job that removes expired records permanently is written but not yet scheduled, so expired rows may persist in our database beyond that window until we finish wiring it. You can email us to have one removed immediately. See our MCP connector lifecycle page for the full operational commitment.

What the connector returns

Everything the connector sends back is retained by the chat host. A tool response carries these categories and no others.

  • Sequence data. The DNA sequence you supplied in the tool call, or the sequence we retrieved on your behalf from a file you provided or from NCBI.
  • Feature annotations. The names, coordinates, strands, types and display colors of the features you supplied or that our annotation engine identified.
  • Computed results. The output you asked for: restriction cut positions, predicted digest fragment sizes, primer sequences and melting temperatures, alignment differences, or a design-check verdict, together with the explanatory text and warnings we generate alongside it.
  • A construct handle and its version metadata. The handle is an opaque identifier for a stored snapshot. Per version we return the version number, name, length in base pairs, topology, a sha256 checksum, a seguid checksum, and whether that version has already been saved into a Plasmole library. A handle minted by a signed-in call is bound to that account: presenting it without that account's credentials does not return the snapshot.
  • A claim URL. When a tool produces a plasmid we return an “Open in Plasmole” link containing a single-use token for the pending plasmid record described above. Whoever holds that link has one opportunity to claim the record. The token appears only inside the URL.

Apart from that single-use claim token, tool responses do not include your email address, your Plasmole account identifier, your team or project identifiers, internal database row identifiers, or any authentication credential or session hand-off code.

Delivery to the chat host

When you use Plasmole through a chat AI, each tool response is delivered to the operator of that host and becomes part of your conversation there. For the ChatGPT app that operator is OpenAI; for Claude it is Anthropic. What that operator does with the conversation, including how long it retains it and what it uses it for, is governed by that operator's terms and privacy policy rather than by this one. Responses that produce a plasmid map include the full sequence, because the map is drawn from the bases themselves. If you do not want a sequence delivered to a chat provider, do not pass it to the connector; use the Plasmole web app instead. The “AI Processing” section below describes the Plasmole web application and its own AI assistant, and does not extend to the connector: there, the model that calls our tools and receives their output is the host's, which we neither select nor control.

Retention of connector records

A construct snapshot holds the sequence, its features, the topology, the name and the two checksums. The stored row additionally records, for our own use rather than for return to the chat host, the account that created it, the connector session it came from, the handle of the version it was derived from, its expiry time, and, once you claim it, the identifier of the plasmid it became in your library. Those linkage fields are what let us show you your own snapshots and stop anyone else resolving them; none of them is returned in a tool response. It expires seven days after the handle is first created, after which the handle stops resolving. As with pending plasmids, the job that permanently deletes expired rows is written but not yet scheduled, so an expired row may remain in our database after it has stopped being reachable. You can email us to have a snapshot removed immediately.

One more connector record, now retired. Between June and September 2026 a tool call made on a signed-in connector also wrote a oauth_handoff_code row: a hashed, single-use, two-minute code that logged you into the web app when you clicked “Open in Plasmole”, stored alongside your Plasmole account identifier and the claim token it pointed at. We removed that hand-off in September 2026: nothing mints one any more, the endpoint that redeemed one is gone, and a tool response no longer carries one. The codes themselves have not been redeemable since then. What we have not done yet is delete the rows already written, and we are being specific about that rather than describing the table as empty. Our nightly cleanup job has a step that deletes them, but the job currently runs in a mode that only reports what it would delete, so those rows are still in our database. Deleting an account removes any that belong to it, and you can email us to have yours removed sooner.

Connected Storage: Google Drive

Google Drive comes with Pro and higher plans, and for every member of a paid lab. We reach nothing in your Drive until you connect a Google account yourself.

What Plasmole can reach

We ask Google for two permissions: drive.file and openid. openid only tells us which Google account you connected. drive.file reaches only the files you pick in Google's file picker. We can't list, search, open or read anything else in your Drive. Google's consent screen calls this “see, edit, create, and delete only the specific Google Drive files you use with this app”. That is Google's wording for per-file access, and Google has no read-only version. Google keeps Plasmole's access to each file you picked until you remove Plasmole in your Google account settings. We never ask for access to your whole Drive.

We keep the connected Google account's email and name to show which account is connected. Disconnecting deletes them.

Your Google Drive folder

Each file you pick goes into your Google Drive folder for that workspace. Only you see your Google Drive folder. Labmates can't open your Drive files or have the assistant read them. To share a file, make a copy in a project, or have labmates add it themselves. The file stays in Google Drive. We keep its name, type, Drive address, last edit time, last editor and Google's version marker, to show and open it. We don't keep its contents, except the text the assistant reads (see below).

Removing a file from your folder changes nothing in Google. We won't open or read it from that folder again unless you add it back. We keep its name, address and last editor, so adding it back restores the same entry.

Seeing a file in Plasmole

Clicking a Drive file shows it inside Plasmole, view only. Plasmole reads it from Google with your access when you open it. Your browser holds it only while the tab is open. We don't store it, and we tell your browser not to save a copy. To edit it, open it in Google.

Making a copy

You choose when to make a copy. A Google Doc becomes a notebook entry. A Sheet, PDF, Word or Excel file becomes a Library file in the project you pick. A Sheet's copy is its Excel export. A PDF, Word or Excel copy stores the file itself, like an upload. Anyone who can see the project sees the copy and its source: the file's name, Drive address and last edit time. The copy keeps that record, plus a fingerprint of what we copied, even after the file leaves your folder. A copy is an ordinary Plasmole file, kept until you delete it. Changes in Drive never reach it. Deleting a copy never touches the original.

What we send back to Drive

Nothing. We never create, change or delete files in your Drive. If that ever changes, this page will change first.

The AI assistant

The assistant uses only the Drive files in your own folder, in the workspace you're chatting in. While answering you, it can list them, find one by name and read it. You can also attach one. It reads the current version from Google, up to four files per message. It reads a Doc's text, every Sheet tab, and the text of PDF, Word (.docx) and Excel (.xlsx) files. For PDF, Word and Excel files, we download the file into memory and keep only its text. From a very large file it reads only the first part, and says so. It doesn't read Sheets, PDFs, Word or Excel files over 10 MB. If AI is off for the workspace, it lists and reads none of them.

We keep the text it read for up to five minutes, for follow-up questions, then delete it. For each read we keep a receipt: version, time, size and a fingerprint, never the contents.

With your message, we send the model provider the listed files' names and edit times, and each read file's text and last editor. That provider is Anthropic, or OpenAI on failover (see “AI Processing”). Both say they don't train on data sent through their APIs. The assistant doesn't automatically add Drive text to its memory of your work.

Its answers are kept with your conversation like any message, including any Drive text they quote.

Drive text, from a file or a copy, is marked untrusted, so the assistant asks before changing your project. A copy is read like any upload, when you add it to a conversation.

Disconnecting

Disconnecting ends our access to your Google account and destroys the stored credential. Your Drive files stop opening, the assistant can't read them, and recently read text is deleted. The folder keeps its entries, but not their names or details. Reconnecting the same account restores the files Google still allows. Another account gets none. Your copies stay, and nothing in your Drive changes. Removing Plasmole in your Google account settings has the same effect.

If your plan stops including Google Drive

New use stops at once: no new files or copies, Drive files stop opening, and the assistant can't read them. Your copies stay. We keep the Google credential for 30 days in case your plan returns, then destroy it.

Google's own rules

Our use of data from Google APIs follows the Google API Services User Data Policy, including its Limited Use rules. We don't use Drive content to create, train or improve any AI or machine learning model. We send it only to the model providers named above, to answer your question.

How We Use Your Information

  • Provide and improve our plasmid design services
  • Offer customer support and respond to your inquiries
  • Send important updates about our platform and services
  • Analyze usage patterns to enhance user experience
  • Ensure platform security and prevent unauthorized access

Data Security

We implement industry-standard security measures to protect your data, including encryption in transit and at rest, secure access controls, and automated dependency and secret scanning in our build pipeline. We have not completed a third-party security audit or a SOC 2 examination, and we do not claim either. We do not sell your research data or disclose it to anyone outside the recipients described in Data Sharing below. Using the AI assistant does send sequence and project context to our model provider; that is described in the section below. For technical detail on how this works in practice (encryption primitives, access control, subprocessor list, and our compliance roadmap), see our Security page.

AI Processing

This section describes the Plasmole web application and its built-in AI assistant. It does not describe the MCP connector, where the model belongs to the chat host; see “Delivery to the chat host” above for that.

We do not train models on your data. Anthropic powers the Plasmole assistant, and OpenAI is configured as a failover that receives the same context when Anthropic is unavailable or rate-limited. OpenAI is more than a failover, though: it also does two jobs on the normal path, because Anthropic offers no product for either. It computes the embeddings that let the assistant search its own memory, so the text of a stored memory reaches OpenAI when it is written, and the text of each message you send the assistant reaches OpenAI so the assistant can search that memory for it; and it transcribes voice input, so a recording you make with a microphone button reaches OpenAI when you stop recording. Those are the only two model providers we send data to from our own servers. Both state that they do not train their models on data submitted through their APIs, and we hold no arrangement permitting it.

Uploading a sequence into the web app does not send it to a model. Our analysis tools are deterministic and run on our own servers. Within the web app, sequence and project context is transmitted to Anthropic, or to OpenAI on failover, only when you use the AI assistant. A recording you make with the microphone in a notebook entry is sent to OpenAI to be transcribed; nothing else from the entry is sent with it. Through the connector, tool responses reach whichever provider operates the chat host you connected, which for the ChatGPT app is OpenAI.

We store your assistant conversations on your account, including your messages, the assistant's responses, and the context you had selected, so the assistant keeps continuity between sessions. The assistant also builds a memory of your work. Those records are scoped to your account, and to your team only when you work inside a team workspace. They are never pooled across customers.

Cookies and Analytics

Our public pages, including the home page, pricing and docs, store nothing on your device. Inside the app, your browser stores your sign-in session and display settings, such as which sidebar sections are collapsed. PostHog, our analytics provider, adds a cookie and a browser-storage entry that link your visits to your account. Sentry keeps a short-lived marker for the browser session so that an error can be matched to its replay. We use PostHog to record which features are used, keyed to your account id; it receives no sequences, record names or research entry content. We use Sentry to capture errors, and a replay of a sampled fraction of sessions for debugging with on-screen text and typed input masked. An error report carries the stack trace, the page or API route and your account id. We remove credentials, email addresses and long sequences from it before it is sent, but it can still include a short sequence, such as a primer, or the name of a file or record the error was about. Our Security page lists what each of them receives. We use Stripe for payment processing on paid plans; Stripe receives your billing email and a tokenized payment method, and we never see your card number. Email us if you want analytics disabled on your account.

Data Sharing

We do not sell or trade your personal information or research data. We share it only as necessary to provide our services, or as required by law. Each provider handles it under its own published terms and data-processing commitments.

The recipients are: our cloud hosting and database providers (Supabase and Render); Resend, which delivers our email; our analytics and error-monitoring providers (PostHog and Sentry); Stripe, for payments on paid plans, which also emails invoices to labs billed by invoice; Anthropic, for the web app's AI assistant; OpenAI, which is the assistant's configured failover, which computes the embeddings that let the assistant search its own memory, so the text of a stored memory is sent there when it is written and the text of each message you send the assistant is sent there to search that memory, and which transcribes voice input, so a recording made with a microphone button, in the assistant or in a notebook entry, is sent there and is not stored by us; the operator of any chat host you connect through the MCP connector (OpenAI for the ChatGPT app, Anthropic for Claude); Google, for sign-in with Google and for the fonts on our pages; the United States National Center for Biotechnology Information (NCBI); and the public research databases named below, when you or the assistant look something up in them.

When you ask the connector to fetch an accession or look up a gene by name, the accession or search term you supplied is sent to NCBI's Entrez service to retrieve the record, together with our shared API credential and contact email. When you run a remote BLAST search in the web app, the query sequence is sent to NCBI's public BLAST servers. NCBI does not receive your account identifier. Workspaces with AI processing turned off cannot run remote BLAST. NCBI's handling of those requests is governed by its own policies.

Other public research databases receive only the identifier or search term a lookup needs: a protein accession or name to UniProt; a gene symbol or identifier to Ensembl; a protein accession to the AlphaFold database at EMBL-EBI; a structure id or keyword to the RCSB Protein Data Bank; a compound name to PubChem, which NCBI runs; a DOI to Crossref, to check a citation; and a plasmid id, part name or search term to Addgene, the iGEM Registry or SynBioHub, when you search or import from them. None of them receives your account identifier, and none of them receives a sequence of yours: remote BLAST, described above, is the only lookup that sends one. Each database’s handling of those requests is governed by its own policies.

When you sign in with Google, Google learns that you are signing in to Plasmole and sends us your name and email address. Our pages load their fonts from Google Fonts, which receives your IP address and browser details when a page loads.

Resend, our email delivery provider, receives the address and content of every email we send, which can include the name of a lab or record and a note or comment someone wrote. Supabase, our authentication provider, sends sign-up confirmation and password-reset emails through it. Our own servers send these through it: invitations to a lab or to a shared record; a request to review a plasmid, and the reviewer’s decision; notices about your access, for example when you are removed from a lab, a lab is deleted or scheduled to be erased, or access to a connected Google Drive ends; a receipt when a workspace is erased; and notices to a lab’s leads when its storage reaches 90% or 100% or is raised. Other notifications reach you only inside Plasmole.

Your Rights

  • Export your plasmids and sequences at any time (GenBank, FASTA, and other standard formats)
  • Request correction of any inaccurate information
  • Delete individual plasmids, entries, and projects yourself, and permanently erase your own items from the trash (in a lab, only its leads can erase the lab’s items)
  • Erase the assistant’s memory of your own work yourself; facts you contributed to a lab stay with the lab, with your name removed
  • Request deletion of your whole account, or of a whole lab workspace, by email
  • Control how your data is shared through project and team permissions
  • Opt out of non-essential communications

Data Retention

We retain your account information and research data for as long as your account is active. Deleting a plasmid, entry, or project moves it to a restorable trash, and you can permanently erase your own items from that trash. In a lab, only its leads can erase the lab’s items. An item that other records still depend on cannot be erased until they no longer do; it stays in the trash until then.

There is no self-serve account-deletion button today. To have your account removed, or a whole lab workspace with it, email us. We confirm the request with you, and a second member of our team then has to approve it; our systems refuse the same account for both. Approval locks the workspace immediately: its members lose access to it, pending invitations expire, live collaboration goes read-only, and the accounts you asked us to close are deactivated, with their API keys and connector authorizations revoked. If an account you want closed is the only admin of another lab that still has members, or owns a record another lab still depends on, we will ask you to hand that over first. You have 72 hours from approval to withdraw the request. Withdrawing it unlocks the workspace, though sessions, API keys and connector authorizations that were revoked stay revoked and have to be set up again.

After those 72 hours the deletion is scheduled for our next nightly job, so it happens no sooner than 72 hours after approval. It permanently removes the workspace’s records, the files uploaded to it, the assistant conversations and the assistant’s memory of that work, and the accounts you listed, including their sign-in records. It cannot be reversed once it has started. We then send a written confirmation that lists what was deleted table by table, a line for every file removed, what was kept and why, and what the deletion does not reach.

Some things outlive the deletion, and the confirmation names them. Billing and payment records are kept for financial record-keeping, Stripe keeps its own record of the payments, and we keep the erase request and its confirmation, including the contact email and workspace name, as the record of the deletion. A file that a record we are keeping still needs is kept with that record. Other labs keep their own records and activity history; where an erased account appears in them, its id is replaced with a marker for an erased user, or removed. Chats in other labs are not searched, and nor are personal chats that mention the erased work without naming one of its records. Data already written to our backups stays there until those backups roll off, on a schedule designed so that every copy is gone within 30 days of approval; the confirmation gives the date by which each backup’s last copy is due to be gone, and says so if the deletion itself ran too late to meet 30 days. Our analytics and error-monitoring providers hold their own copies: we file a deletion request with each and record those requests on the confirmation. Our model providers keep their own API logs under their own terms, which we cannot delete for you. Some records may be retained where required by law.

Changes to This Policy

We may update this Privacy Policy from time to time. We will notify you of any significant changes via email or through our platform. Your continued use of Plasmole after such changes constitutes acceptance of the updated policy.

Contact Us

Plasmole is operated by Plasmole, Inc., a Delaware corporation, United States. Plasmole, Inc. is the party responsible for the data described in this policy.

If you have any questions about this Privacy Policy or how we handle your data, including a request to access, correct or delete your data, please contact us at:

[email protected]

Discover the future of molecular biology.

Start freeBook a demo
Plasmole
Product
  • Product overview
  • Plasmole AI
  • Chat tools (MCP)
  • Integrations
  • Pricing
  • Changelog
  • Documentation
Solutions
  • Academic labs
  • Biotech teams
  • Contract research
Company
  • About
  • Security
  • Compare
  • Contact
Social
  • LinkedIn (opens in a new tab)
  • Slack (opens in a new tab)
© 2026 Plasmole, Inc.
PrivacyTerms