Melioda

Privacy Policy

Last updated 3 August 2026

This policy explains how Groundery handles information in Melioda, our AI post-production platform. It covers two different things: the personal data we hold about you as a customer or visitor, and the material you upload for processing — which we treat as yours throughout. It applies to melioda.com, the Melioda application, and the emails we send about them.

01The short version

  • We never train on your content.Not our models, not our vendors’ models. No exceptions, no opt-out to find in a settings page.
  • Your uploads stay private. Media sits in a private bucket and is served only through short-lived signed links, scoped to your workspace.
  • We don’t sell or share personal data for advertising, and we run no advertising or cross-site tracking cookies.
  • You can get it out, and get it deleted. Export any time; deletion on request, and after termination on the schedule in clause 8.

02Our two roles

For account, website, support and marketing data, Groundery is the controller: we decide why and how it is processed, and this policy is the notice for it.

For the material you upload and everything derived from it (“Customer Content”), Groundery is a processor: you decide what goes in and why, and we process it on your instructions to run the service. Where that content contains personal data — the voices, faces and names of the people in your recordings — you are the controller, and you are responsible for the notices and consents described in clause 4 of the Terms. A data processing agreement is available on request.

03What we collect

Account data. Name, work email, organization, workspace role, and authentication metadata (sign-in times, method, IP address of the session), handled through Clerk. We never see or store your password.

Access request data. If you use the request form: your name, work email, organization, and what you told us you publish. We use it to evaluate and respond to the request, and to contact you about a pilot.

Customer Content. The audio, video, transcripts, documents and brand assets you upload, and everything the pipeline derives from them — transcripts with word timings, speaker labels, selected moments, research summaries with their sources and images, rendered clips, audiograms, articles and story packages.

Usage data. Pages viewed, features used, content hours processed, job outcomes and errors, plus standard server logs (IP address, user agent, timestamps). Product analytics run through Vercel Analytics, which is cookieless and does not build a cross-site profile of you.

Communications. Emails and messages you send us, and our replies.

04Why we process it

  • To provide the service: transcribe, analyze, research, repackage, store and deliver your material, and to keep your workspace working (legal basis: performance of a contract).
  • To secure the service: detect abuse, investigate incidents, enforce usage limits (legitimate interests).
  • To support you: answer questions, diagnose failed jobs, and — only with your permission for the specific case — look at a session to debug it (legitimate interests, your request).
  • To improve the service using aggregate operational signals such as failure rates, durations and cost per hour — never the content itself (legitimate interests).
  • To bill and account for usage, and to meet tax and accounting obligations (contract, legal obligation).
  • To send service messages, and — separately, and only if you asked for them — product updates you can unsubscribe from at any time (contract, consent).

05Model training

Customer Content is never used to train, fine-tune, or evaluate any machine-learning model — ours or a vendor’s. Our AI subprocessors are engaged under terms that prohibit training on data we send them, and we use zero-retention processing where a provider offers it.

What we do learn from is operational metadata: how long a step took, whether it failed, how many content hours a workspace processed. That tells us where the pipeline is slow or brittle without anyone reading your material.

06Automated processing of the people in your recordings

Running the pipeline means analyzing the people who appear in what you upload. Specifically, Melioda separates speakers and attributes lines to them; labels a speaker by name where the recording itself makes the name available; samples frames and asks a vision model how many people are visible together, so a clip can be framed or cropped correctly; and looks up public figures and organizations mentioned in the material against open-web sources.

We do not create biometric templates, do not perform facial recognition against any identity database, and do not match people across recordings or across workspaces. Frame analysis counts and locates people in a shot; it does not identify them.

07Who else processes it

We use the vendors below, each under a written agreement limiting them to processing on our instructions. We do not sell personal data, and we do not disclose it to anyone else except where the law requires it — in which case we will tell you unless we are barred from doing so.

VercelApplication hosting, delivery, and product analytics (United States).
ClerkAuthentication, user accounts, and session management.
NeonManaged Postgres — workspace, session, transcript, and research records.
Cloudflare R2Object storage for uploaded media and rendered assets. Buckets are private; access is by short-lived signed URL.
InngestQueues and runs the processing pipeline.
DeepgramSpeech-to-text: transcripts, word timings, and speaker separation.
OpenAIAnalysis of the transcript — moment selection, structuring, and drafting of repackaged assets.
AnthropicAnalysis and drafting, for the steps run on Claude models.
Google (Gemini)Vision analysis of sampled frames for framing and cropping, and on-demand image generation.
Perplexity, TavilyOpen-web research lookups for the names, organizations, and claims found in your material.
ResendTransactional email, including access requests and notifications.

Research lookups are the one step that reaches outside this list: to find context for a name or a claim, the research provider queries the open web and returns pages, sources and images from third-party sites. The query carries the entity and a short piece of surrounding context from your material — never the whole recording or transcript.

We will give notice before adding a subprocessor that processes Customer Content, and this list is the record of who they are.

08How long we keep it

  • Customer Content: for as long as your workspace is active. Delete a session and it is removed from live systems within 30 days, including its media, renders and derived records.
  • After termination: 30 days to export, then deletion from live systems within a further 30 days. Encrypted backups cycle out within 90 days of that.
  • Account data: for the life of the account, then up to 12 months after closure for security and dispute-resolution purposes.
  • Access requests we don't take forward: up to 24 months, so we know we already spoke.
  • Server and job logs: up to 12 months.
  • Billing and tax records: as long as tax and accounting law requires.

09Security

Data is encrypted in transit (TLS) and at rest. Media lives in private object storage and is reachable only through signed URLs that expire; nothing is served from a public bucket. Access is scoped to a workspace and enforced on every request. Internal access to production is limited to the people who need it, is logged, and is used for operating the service — not for reading your material. No system is perfectly secure; if a breach affects your personal data we will notify you and, where required, the relevant authority without undue delay.

10Where data is processed

Melioda runs on infrastructure in the United States, and the vendors in clause 7 process data there and, in some cases, in other countries where they operate. Where personal data protected by EU or UK law is transferred out of those regions, the transfer is covered by the European Commission’s Standard Contractual Clauses (with the UK Addendum where applicable) or another lawful transfer mechanism.

11Your rights

Depending on where you live, you may have the right to access, correct, delete, or receive a portable copy of your personal data; to object to or restrict certain processing; and to withdraw consent you previously gave. If you are in the EEA or UK you also have the right to complain to your supervisory authority. If you are in California, you have the rights to know, delete, and correct, and the right not to be discriminated against for exercising them — and note that we do not sell or share personal information as those terms are defined by the CCPA/CPRA, and have not in the preceding twelve months.

Write to hi@groundery.com and we will respond within 30 days. If your request concerns material uploaded by one of our customers — for example you were recorded in an interview — we will refer you to that customer, who controls it, and support them in responding.

12Cookies

We set only what the product needs: session and authentication cookies from Clerk that keep you signed in and protect against request forgery, and a small number of preference cookies. Product analytics are cookieless. We run no advertising cookies, no third-party trackers, and no cross-site profiling, so there is no consent banner to dismiss.

13Children

Melioda is a business tool and is not directed to children. We do not knowingly collect personal data from anyone under 18 as a user of the service. Recordings you upload may of course feature people of any age; handling that lawfully is your responsibility as controller of that material.

14Changes to this policy

We will post any change here and move the “last updated” date. For material changes affecting how Customer Content is processed, we will give at least 30 days’ notice to workspace administrators by email before it takes effect.

15Contact

Groundery — privacy questions, data requests, and data processing agreements: hi@groundery.com.