Privacy Policy
InkVell Inc. dba Synthetic Sciences ("Synthetic Sciences," "we," "us," or "our") provides OpenScience, a scientific research workbench; hosted research graphs, including services previously called Atlas; and OpenScience Ace, our managed model-access and billing service. This Privacy Policy covers personal information we handle through these services, our websites, accounts, APIs, and support channels (the "Services").
This policy explains our practices; it is not a request for consent to every use described here. Our Terms of Service govern use of the Services. A separate agreement with your organization, including an applicable data processing agreement, may govern information we process on its behalf.
1. Who is responsible for your information
InkVell Inc. is responsible for the account, website, billing, security, and other processing described in this policy. If an employer, university, or other organization provides your account or workspace, it may separately control the research information submitted through that workspace. Ask its administrator about its instructions, access rules, and retention requirements. Where we act as its processor or service provider, we handle that information under the applicable agreement and assist it with privacy requests.
This policy does not govern a model provider, website, repository, plugin, or other service you use independently. Their own terms and privacy notices apply to their processing. Using open-source software solely on infrastructure you control does not, by itself, give us access to all of that infrastructure or its files.
2. Information we collect and its sources
- Account and identity information: your name, email, profile details, authentication identifiers, workspace memberships, roles, invitations, and account or API-key activity. Sign-in providers such as Google or GitHub supply information associated with the sign-in method you choose.
- Research content: prompts, conversation context, model responses and provider-visible reasoning, files and images you submit, code, commands, tool inputs and results, search queries, research notes, graph nodes and relationships, artifacts, and associated project metadata. This can include unpublished work and personal information about you or others.
- Connection information: provider keys, OAuth tokens, repository permissions, and service credentials when you save or connect them to a hosted workspace. Device-only credentials and credentials deliberately shared with a workspace have different storage and access paths.
- Payment and transaction information: purchased Wallet balance, promotional balance, usage, charges, payment status, invoices, billing details, and payment-method references or limited card descriptors received from Stripe. Card details entered into Stripe checkout are handled by Stripe; we do not need your full card number or security code for our usage ledger.
- Technical and activity information: IP address and connection details, browser or device characteristics, application version, language and time zone, page visits, request and session identifiers, model selections, token counts, latency, errors, and security events. What we receive depends on the service and software version.
- Communications: messages, feedback, files, and diagnostic material you send to support, along with contact and delivery information. We also receive information from collaborators, administrators, payment providers, and connected services as needed for the features you use.
3. Local work, hosted graphs, and connected services
OpenScience stores conversations, project state, local traces, logs, settings, and credentials on the machine running the workbench. Signing out does not delete these files. A remote workbench stores them on its host. Local storage does not mean every operation stays offline: model requests, searches, tools, repository connections, and other integrations can send selected content to external services.
On direct routes, your own provider key, supported provider sign-in, local model, or custom endpoint sends model requests to that selected service without using the Ace inference gateway. Separate managed features and trace uploads in earlier versions can still send information to us. Changing the model route alone does not change every data-sharing or billing setting.
Content you upload to hosted graphs is stored by our service. Graph search and indexing can send graph text or search queries to an embedding provider through OpenRouter. Private access controls restrict who can retrieve a graph; they do not mean the content is never processed by hosting or model providers.
Collaborators can access content according to the permissions you grant. Where public sharing is available and you enable it, other people may read, copy, or redistribute the shared material. Revoking access cannot recall copies already obtained. A downloaded artifact or signed download link may be accessible to anyone who possesses it while it remains valid.
4. Provider keys, OAuth tokens, and account deletion
Provider API keys, OAuth access and refresh tokens, and other secrets saved in our hosted credential store are encrypted at rest and accessed to authenticate connections you authorize. They must remain usable for those connections, so encryption is not anonymization. Credentials stored only on your workbench have their own local storage protections; this hosted-storage statement does not mean every file on your device is encrypted.
The trace pipeline attempts to remove or mask recognized credentials before storing research traces. This redaction is intended to keep secrets out of logs and downstream research datasets, but it cannot guarantee that every secret in arbitrary text is detected. Credential values belong in connection settings, not in prompts, feedback, or support messages. Redacting a credential does not anonymize the surrounding research or revoke the original credential.
To delete your account, email team@syntheticsciences.ai from your account address. When we complete a verified account-deletion request, we delete the personal provider keys, OAuth access and refresh tokens, and other connection secrets we hold for that account from active hosted credential stores, and revoke Synthetic Sciences API and device access keys issued for that account. This is a request-based process; signing out, uninstalling the app, or deleting research traces alone does not perform it. We explain any legally required retention or other applicable limits when responding.
Deleting our copy of a third-party token does not necessarily revoke it at its issuer, disconnect every other application, or delete your provider account. Revoke the connection or rotate the key in the provider’s own account settings to invalidate it there. Account deletion does not remotely erase credentials on your devices, self-hosted workbenches, external repositories, or backups you control. Remove those copies separately, including copies previously received through workspace sharing.
For organization-owned connections, deleting a member’s account removes that member’s authorized service access; it does not itself require deleting the organization’s credentials or other members’ connections. Organization administrators should rotate shared credentials when needed because previously delivered copies cannot be recalled. Retained revocation and security records may include identifiers or key fingerprints without retaining a usable secret.
The active-store deletion commitment is subject to the backup and legal-preservation limits in Retention and the limits of deletion. We protect retained copies and limit their use; we do not promise immediate erasure from every backup or from an independent provider. Contact us if you need deletion of a particular connection without closing your account.
5. Ace requests and required service records
Ace receives the request content needed to perform managed inference, including prompts, conversation history, attachments, tool definitions, and tool results included in the request. We pass that content to OpenRouter and the model provider serving the request, or the provider identified for the managed feature. The resulting output passes back through the service. Provider routing and retention depend on the selected model, endpoint, and applicable provider terms.
We retain operational and financial records to deliver requests, measure usage, settle Wallet charges, prevent duplicate charges, investigate failures or abuse, and meet accounting obligations. These records can identify your account and funding workspace and include provider/model, request identifiers, token counts, cost, timing, and outcome.
Some non-streaming model responses and search results are stored so the service can replay a completed operation safely after a retry. These records can contain research content. Turning off optional research sharing does not disable the processing needed to carry out a managed request or erase required financial and security records. Ace is not offered as a blanket zero-retention service.
6. Research traces and logging
A research trace is a record of activity across a session. It can include prompts, project context, assistant output, reasoning returned by a provider, model and route identifiers, tool calls and results, commands, searches, artifact information, errors, retries, and timing. Material from a private file can be included when a tool reads it into the session. This is content collection, not merely anonymous traffic measurement.
Trace behavior varies by installed version. Earlier OpenScience releases enable uploads after account sign-in by default. Some versions offer a "Use my data" control; later versions of that uploader label managed Ace traces "Always on" and offer a separate "Share user-owned routes" control for BYOK, provider-subscription, local, and custom routes. Disabling that latter control stops sharing from those routes; it does not stop managed Ace tracing. Current local-workbench versions have removed that uploader and its settings. Previously uploaded records and the hosted ingestion service are separate from the version installed on your device.
Uploaded traces are associated with account and installation identifiers and may be attributed to a workspace. Authorized personnel can inspect retained content for permitted operational purposes through restricted administrative tools. They are not anonymous merely because identifiers are hashed or credentials are masked.
The trace pipeline attempts to redact recognized API keys, passwords, tokens, private keys, and credential-bearing URLs. It does not reliably remove all personal information, confidential research, or secrets in arbitrary text. Do not treat automated redaction as permission to submit restricted data. The trace preference also does not control logs held by a provider, connected tool, or your own machine.
7. Model training, evaluation, and improvement
We use research content and usage information to evaluate and improve the research experience and to train or improve models. This can involve reviewing examples, investigating failures, measuring quality, and preparing training or evaluation datasets from content collected through the Services. This use can involve unpublished research; it is not limited to feedback messages or public material.
We seek to reduce identifying information when preparing data for training, but we do not promise that research content is anonymous. Source records may still be linked to your account, and research text can itself identify a person or project. Pseudonymized or otherwise identifiable information remains subject to this policy and applicable privacy rights. We do not treat simple credential masking as proof of anonymization.
Applicable data-sharing choices, organizational agreements, and legal requirements govern this use. Where consent is required, we must obtain it before processing for that purpose; accepting the Terms or signing in is not a substitute for that consent. Contact us to request exclusion of identifiable research content from future training or evaluation and to discuss restrictions required by your organization.
Organizations can contact us to arrange separate terms prohibiting retention of research content and its use for training. Do not assume those terms apply merely because you create or join a workspace: the scope, eligible features, provider handling, and activation must be confirmed in writing before submitting content under that arrangement. Necessary account, payment, and legal records must be addressed separately.
Stopping future collection does not automatically delete an existing training dataset or reverse a completed training run. Requests involving identifiable source records or derived datasets are reviewed under applicable law. Data that has been irreversibly anonymized may no longer be attributable to an account. We do not promise that removing a source record removes its influence from an already trained model.
Model providers have separate practices. A direct provider account and an Ace request may have different terms even when the model name is the same. Review the provider terms for your route, including OpenRouter’s privacy policy and provider logging guidance. Our data-sharing controls do not change settings in an independent provider account.
8. Other purposes and legal bases
We use information to operate accounts and workspaces, authenticate access, provide research and collaboration features, process payments, respond to support, communicate service changes, secure the Services, prevent fraud and misuse, comply with legal requirements, and resolve disputes. We use traffic and activity information to understand reliability and feature use. Training and research improvement are described separately above.
Where a law requires a legal basis, we rely on performance of our contract for requested account, research, and payment services; legal obligations for required financial records and lawful disclosures; and legitimate interests, where permitted and balanced against your rights, for security, fraud prevention, service reliability, and support. We rely on consent for processing that requires it, including applicable optional content uses or tracking. You may withdraw consent without affecting the lawfulness of processing before withdrawal.
We may produce aggregate statistics and de-identified information. We maintain information that is truly de-identified in that form and do not attempt to re-identify it except for legally permitted testing of the de-identification process. Information that can still reasonably be linked to a person is not excluded from this policy.
9. Who receives information
- Infrastructure and account providers: hosting, database, storage, authentication, email, and operational vendors process information needed to provide their services. Integrations include Supabase for hosted accounts and storage, Vercel for the website and analytics, and infrastructure hosting such as Fly.io.
- Payments: Stripe processes checkout, saved payment methods, transactions, and payment verification.
- Research providers: OpenRouter and the selected model providers process managed inference and embeddings. Managed web research can send search queries or URLs to providers such as Firecrawl or Exa. Connected tools and repositories receive information needed for the tasks you request.
- People you authorize: collaborators and workspace administrators receive shared content, membership or account information, and usage information according to their roles. Workspace administration does not automatically grant access to every private file on a member’s device.
- Personnel and advisers: authorized employees, contractors, and professional advisers may access information where needed for the purposes in this policy and subject to appropriate confidentiality obligations.
- Legal and business disclosures: we may disclose information when required by applicable law or valid legal process, or when reasonably necessary to address fraud, security threats, or serious harm, protect legal rights, or enforce our agreements. In a merger, acquisition, restructuring, or asset transfer, information may be disclosed with appropriate safeguards and any required notice.
We do not sell personal information or share it for cross-context behavioral advertising. We do not grant advertisers access to private research traces. Listing a provider here does not mean every request is sent to every provider. Contact us for information about the providers involved in your service or for an organizational processing agreement.
10. Website analytics and browser storage
We store your cookie choice for up to 180 days and remember the theme you choose in browser local storage. You can change it at any time through Cookies in the footer.
When you allow Analytics, the landing website uses Vercel Web Analytics and Google Analytics to measure page visits, referring sites, coarse device or location information, and useful actions such as contact and product-link clicks. Marketing events omit form text, email addresses and research content. Vercel describes its analytics service as using no cookies to identify visitors across browsing sessions.
If we add personalized advertising or business visitor identification, the cookie notice will name it and ask again before it starts. Accepting would then permit Google advertising measurement and audience creation, or let the provider named in the notice associate eligible visits with companies or professional profiles. A Global Privacy Control signal disables advertising and visitor identification.
Withdrawing a choice stops future collection by the relevant website tags and clears the matching first-party cookies we control. Records already sent to a provider are subject to its retention and deletion process. Contact us for deletion requests. Each of our websites stores its own choice.
Hosted infrastructure receives connection information when you visit. Signed-in applications use separate storage for authentication and settings. The website choices described here govern marketing-site activity; product data controls are described in their respective sections.
11. Retention and the limits of deletion
For standard accounts, retained research content and training or evaluation datasets have no preset deletion date. We retain them on an ongoing basis for the purposes described here unless you request deletion, a separate agreement sets a limit, or applicable law requires deletion or a shorter period. Retention without a preset date does not override legal requirements to keep personal information only as long as justified by its purpose. Different stores have different lifecycles; a single trace-retention period does not apply to all research, billing, or training data.
- Local research and credentials remain on your machine until removed through the application or your own storage controls. Account deletion does not erase your devices, external repositories, or backups you control.
- The raw uploaded-trace store normally assigns a 30-day expiry and uses scheduled deletion. Completion can depend on the processing cycle. This is not a 30-day limit on separately retained research copies, training datasets, trace-derived usage records, consent history, or deletion records.
- Replayable API-response records normally have a seven-day expiry and scheduled cleanup. Hosted graph content, files, and search records have separate lifecycles and can remain while needed to operate those features, provide access, or investigate a request. An expired cache or replay window is not a promise that every underlying copy has been deleted.
- Account, invoice, Wallet, payment, fraud, and security records may be retained after an account closes where required for accounting, legal obligations, dispute resolution, or security. A request to delete research traces does not reverse charges or erase required financial records.
- Retained research copies and training or evaluation datasets have no automatic expiry under the standard arrangement. You can request deletion or exclusion from future training by email. We assess the request for the relevant source records and identifiable derived datasets, subject to legal exceptions; the raw-trace expiry does not set a deletion deadline for datasets or trained models.
- Backup copies and material subject to a legal preservation requirement may remain after active records are removed. Retained personal information remains protected and subject to limits on its use. Independent recipients may have their own retention obligations.
Contact us for the retention terms applicable to your account or a specific category of information. We assess deletion requests across the relevant stores rather than treating a trace deletion as deletion of all information about you.
12. Your choices and deletion requests
- Choose the model route and connected services appropriate for your work. Direct BYOK or local inference changes where model requests go; it does not disable separately requested hosted features or historical trace sharing.
- If your installed version has data-sharing controls, review their scope before relying on them. Turning sharing off affects future uploads after the preference is applied. It is distinct from requesting deletion of previously received data. If the setting is missing, contact us to record your request.
- Ask us to access, correct, export, restrict, or delete personal information, to close your account, or to exclude identifiable content from future training and evaluation. State which request you intend: account closure, trace deletion, training exclusion, and payment cancellation are different actions.
- Disconnect integrations and revoke credentials in the relevant application or provider account. Our credential-deletion process covers personal hosted credential copies when account deletion is completed; local copies and provider-side revocation require separate action.
- Manage future Ace automatic reloads in billing. Signing out, uninstalling OpenScience, requesting research-data deletion, or switching model routes does not by itself cancel that authorization.
Send requests to team@syntheticsciences.ai. Include your account email and the type of request; do not send passwords, API keys, or an entire private research dataset. We may request proportionate information to verify your identity or authority. If an organization controls the information, we may direct you to its administrator and help it respond.
13. Regional privacy rights
Depending on your location and the law’s applicability, you may have rights to know what information we collect and disclose, access it, receive a portable copy, correct it, request deletion, restrict processing, object to processing based on legitimate interests, or withdraw consent. You may also have a right to limit certain uses of sensitive information, use an authorized agent, and appeal a denied request. We do not discriminate against you for exercising applicable privacy rights.
For California residents, the categories described above include identifiers; commercial information; internet, device, and network activity; professional or educational details you provide; and the contents of research submissions and communications. Account credentials and certain content may be sensitive personal information. The sources, purposes, recipient categories, and retention criteria in this policy apply to those categories. We do not sell or share personal information for cross-context behavioral advertising, including information about minors.
For people in the EEA, UK, or Switzerland, you may object to processing based on legitimate interests and complain to your local data protection authority. People in other locations may also contact their competent privacy regulator. You do not need to contact us first to exercise a right to complain to a regulator.
We respond within the periods required by applicable law and explain any permitted extension, limitation, or refusal. To appeal a decision where an appeal right applies, reply to our response or email us with the subject "Privacy appeal." Where a right does not apply to irreversibly anonymous information or a legal exception requires retention, we will explain that distinction where relevant.
14. International processing
We operate from the United States and use providers that may process information there or in other countries. Those countries may have different privacy laws from your own. We do not promise that data remains in a particular country unless a separate agreement expressly provides for it.
Where required for a transfer, we use a permitted legal mechanism and safeguards, such as an applicable adequacy decision or contractual protections. Contact us for information about the destinations and safeguards applicable to your service and, where available, a copy of those safeguards. This policy does not itself establish a data processing agreement or certify participation in a transfer framework.
15. Security and sensitive research
We use access controls and other technical and organizational measures intended to protect information. Hosted research and administrative access are subject to authentication and authorization. No system is completely secure, and we do not promise that all content is end-to-end encrypted or inaccessible to service operators.
Research may contain personal health information, human-subject data, confidential business information, or restricted material. Do not submit information unless you have the necessary authority and the selected service is appropriate for its sensitivity. Contact us before using hosted Services for data requiring a special processing agreement or regulated safeguards. A scientific use case does not automatically make a service suitable for clinical or regulated data.
16. Children
The Services are not directed to children under 13 or a higher minimum age required by applicable law. Users under 18 must have a parent or legal guardian’s permission and must meet any higher age requirement of the selected provider. We do not knowingly collect information from children below the applicable minimum age. Contact us if you believe this has occurred so we can investigate and take appropriate action.
17. Changes to this policy
We update the date above when this policy changes. We provide additional notice of material changes, such as through the Services or account email, as required by law. If a new purpose requires consent, we obtain that consent before proceeding. A later policy update does not retroactively authorize a use of previously collected information that requires separate permission.
18. Contact
Send privacy questions, requests, or concerns to:
InkVell Inc. dba Synthetic Sciences
2261 Market Street PMB 27364
San Francisco, CA 94114, United States
team@syntheticsciences.ai