Privacy policy

What Mindcomb stores

• Your account and profile. Your sign-in identity (an email address or configured social login, held by our authentication provider) and your public profile: handle, display name, avatar and bio.
• Runs you ask while signed in, including runs you never save: the prompt, the models you compared, each model's answer and provider metadata (provider, model version, token counts, cost, timing, and how the answer finished or failed). Unsaved runs stay private to you.
• Combs you save: title, category, topic tags, visibility, your pick and its reason tags. Topic tags are visible to anyone who can read the comb; unlisted tags stay link-only. Adding tags does not change your consent choices. New direct comparisons start with Public selected; choose Private before or after saving for owner-only access, and change it any time. Making a comb private stops future access; it cannot recall copies people already saw, copied or shared. Forks other people made keep showing without credit to it, and its upvotes and comments are kept but hidden until you share it again. Sexual content and other comparisons that can’t be shared yet are saved privately. Public combs are listed on Mindcomb; anyone with an unlisted link can open it. Private combs and private-only results never enter public leaderboards or licensed datasets.
• Your private picks and usage signals, while personalization is on: see what personalization records below.
• Saved setups and system instructions. Private to you, including on combs you share.
• Your provider API keys, encrypted on our server. They are never shown again and are used only to call the provider you chose for your own requests.
• Reports you file about combs.
• Upvotes you give. Private: others see only a comb's total, never who upvoted it.
• Your private notification inbox. Anonymous upvote totals, confirmed billing changes and which notices you have read. Notifications never identify voters or reveal Following relationships. They are not personalization or research-licensing inputs. Your account export includes this state.
• Previously saved follow relationships. Following is unavailable in this release. Existing relationships remain private and appear only in the account owner's export. They are not used for notifications, AI profiles, analytics or licensed datasets. Deleting either account removes the relation.
• Image presets and discovery preferences you save.
• Rate-limit records: short-lived counters keyed by your IP address or account, to prevent abuse.
• Your data choices below, with the policy version and time of each change.

On our servers, anonymous use stores rate-limit counters rather than prompts or answers. Sign-in cookies keep you signed in; Mindcomb uses no advertising or analytics cookies. Your theme preference stays in your browser.

For draft recovery, this tab stores your next prompt, draft instructions, selected models, category, setup reference and opaque comparison/source references. Recovery is limited to a 24-hour window; tab drafts are cleared on sign-out or account change. API keys, tokens, answer text and private verdict notes are excluded. Recorded answers are restored through account-authorized reads. Bringing an anonymous draft through sign-in requires a Restore draft choice; recovery never automatically generates, saves or publishes.

Your prompts and system instructions are sent to the model providers you choose (OpenAI, Anthropic, Google, xAI): with your own API key when you are signed in, or, where an operator has enabled an anonymous demo, with Mindcomb's key. Those providers' terms and privacy policies apply to that processing.

Your choices

You can change both choices at any time.

Open Data & privacy settings

Personalize Mindcomb with my history

Mindcomb may use your own runs and picks to tailor features for you, such as which models to suggest for a task. The result is visible only to you. Your picks and run details (never prompts or answers) may also be used, except for private saved or private-only content, without your identity, in internal checks of how well those suggestions work. On unless you turn it off.

These signals come only from the signed-in Hive on Mindcomb, never from embeds. Turning personalization off stops recording them. Picks and notes already recorded remain in your export until account deletion or an eligible unsaved run is explicitly expired under the proposed lifecycle below. Usage signals follow their separate retention rule.

Your AI profile

While personalization is on, Mindcomb works out which models win for you, overall and per task, from your own picks on runs that compared two or more models, with recent picks counting most (their weight halves every 90 days). It is worked out each time you open your overview or the Hive and is never stored; only you see it, and it is never part of a licensed dataset. The Hive uses it to suggest models for your next prompt and records whether you kept the suggestion. To check that suggestions help, Mindcomb occasionally replays the picks of everyone with personalization on in an internal evaluation that reports only totals across many people; it carries no names, emails or account ids, though anyone with full database access could still match the runs to accounts. Pausing personalization stops it at once; a Hive tab already open follows when you next return to it, within seconds. Resetting it on your overview makes only later picks count; your runs and picks are kept, and the reset date is part of your export.

Include my saved combs in licensed research datasets

Mindcomb may license redacted judgments from your saved combs to researchers and model builders. Off unless you opt in. It applies only to exports made after you opt in; turning it off stops every later export. Only combs you saved (public or unlisted) can be included, never forks (remixes) of any comb, including your own, and never follow-up conversations.

Licensed datasets are described as redacted, not anonymous. If this policy changes what licensing may share, the version number changes and an earlier opt-in no longer applies until you opt in again.

Export and deletion

You can download a copy of your data at any time: your profile, combs, runs and answers, verdicts, private picks and notes, usage signals, setups and instructions, which providers have a saved key (never the keys), reports you filed, upvotes you gave, image presets, discovery preferences, your consent history and when you last reset your AI profile and your private onboarding progress.

Export your data

Data & privacy includes a self-serve deletion workflow, kept disabled until hosted integration and operator policies are validated. When enabled, you review a preview, may export first, verify a recent sign-in and explicitly confirm removal. Private saved bodies are erased. Required structural references retain content-free tombstones. Previously shared snapshots retain their published text and model provenance without profile attribution; published text may still identify you. Previously delivered copies cannot be recalled. Private instructions, unsaved evidence, notes, activity and authored comment text are erased. Required lineage and restricted moderation/audit evidence remain. Ask the operator to review removal of retained publications before deleting your account. Contact privacy@mindcomb.ai (a placeholder address until launch) from your sign-in address. Provider-held data, offline copies and backups follow separate policies still pending review.

Text safety screening

In the restricted staging trial, complete prompts, setup instructions, included conversation context and buffered model answers are processed by OpenAI's moderation API using a separate operator credential. This applies even when you choose another generation provider. They are not written to application logs. Your generation provider also applies its own policies.

Only named adult testers are admitted to the restricted text policies. The free checker does not establish consent or reliably cover exploitation; a private-only result is not a certification of consent. Earlier private-trial evidence cannot approve publication. A separately enabled publication beta reuses matching generation checks for eligible direct comparisons, including the exact public prompt, default title, answer and model metadata; Save adds no AI review. Custom titles/tags and unsupported histories stay private. English text and code are the initial evaluation target; there is no automatic language detection or claim of complete coverage.

Retention

Notification history is available for 90 days. Older notices are excluded from reads and export and removed by explicit operator maintenance; physical removal depends on maintenance execution. Minimal per-comb high-water marks remain while the account and comb exist to prevent repeated vote toggles from generating alerts. Account deletion removes notification history and suppression state.

Usage signals (copies and comparison opens) are removed periodically once they are older than 180 days, a placeholder until launch. The proposed lifecycle also makes unsaved private text eligible after 30 days without recorded activity. Saved private, public, unlisted and hidden comparisons and their required ancestry are protected. Explicit operator expiry erases private prompts, answers, instructions, frozen context and judgment notes, retaining content-free request, provenance and spending records to prevent duplicate paid work. Account deletion retains its separate behavior. This cleanup is disabled until hosted scheduling, legal/disclosure, recovery and backup policies are validated; it is not a guaranteed deletion deadline. Provider-held data, downloaded/offline copies, replicas and backups follow separate policies.