Version: v1.3 Streamlined
Operator: Bobby Built Ventures LLC d/b/a QuillCaddie
Applies to: All users, customers, workspace admins, and anyone using QuillCaddie AI, transcription, transcript correction, vocabulary, BYOK, connector, or MCP features.
Quick summary
- QuillCaddie uses AI to distill your captures into useful outputs.
- Your raw source text/transcript remains the source of truth.
- Customer Content is not used for generalized model training by default.
- Business workspaces are training locked off by default unless an authorized workflow or signed agreement says otherwise.
- Operational processing, such as transcription and AI output generation, is not model training.
- Optional model-contribution uses require a clear opt-in.
1. Core principles
QuillCaddie is an AI-assisted idea-capture workspace. AI is a distiller, not the source of truth. The Service is designed around these principles:
- preserve source text/transcripts separately from AI outputs;
- separate explicit points from inferred possibilities, assumptions, questions, risks, and next actions;
- avoid inventing facts or turning speculation into commitments;
- disclose source limitations and transcript quality issues where relevant;
- keep raw audio temporary by default;
- keep Customer Content out of generalized model training by default; and
- give businesses admin controls appropriate for organization-managed workspaces.
2. AI features covered
This Policy covers transcription, speaker diarization where available, transcript correction, Builder Vocabulary normalization, AI distillation, titles, summaries, tags, folder/product suggestions, generated prompts, custom templates, research mode, vision/image processing where enabled, meeting/interview processing, connector-generated artifacts, and MCP/API tools.
3. Customer Content
Customer Content includes captures, raw audio, transcripts, source text, cleaned transcripts, correction metadata, images, screenshots, files, links, notes, prompts, templates, vocabulary, product/project context, connector data, AI outputs, generated artifacts, exports, and related metadata.
4. Operational processing is not model training
You authorize QuillCaddie to process Customer Content to provide requested features, including transcription, AI outputs, search, storage, export, sharing, support with permission, security, abuse prevention, usage/billing, connector actions, MCP/API tools, workspace-specific vocabulary/templates, and legal compliance.
This operational processing is not generalized model training.
5. Default no-training posture
QuillCaddie does not use Customer Content for generalized QuillCaddie model training, fine-tuning, ASR training, source-content datasets, or provider model-improvement programs by default.
Team, Business, Enterprise Preview, and Enterprise workspaces are no-training by default unless an authorized admin/legal workflow or signed agreement enables a specific opt-in.
6. Provider processing
QuillCaddie may send Customer Content to AI and transcription providers to provide requested features. Where QuillCaddie controls provider settings, QuillCaddie should configure accounts to avoid provider model-improvement use of Customer Content. Provider lists and routing information are maintained at Subprocessors and AI Providers.
7. What QuillCaddie may improve without Customer Content training
QuillCaddie may improve the product using:
- content-free analytics and usage metrics;
- latency, reliability, and cost metrics;
- synthetic examples;
- internal founder/team dogfood examples;
- public or non-sensitive examples;
- aggregate statistics that do not identify users or reveal Customer Content;
- user feedback that does not include sensitive content; and
- bug reports or support examples where the user chooses to provide them.
8. Optional model-contribution opt-in
If QuillCaddie offers optional model-contribution settings, the user or authorized workspace administrator must opt in to a specific category before selected Customer Content becomes eligible for internal eval datasets, fine-tuning, ASR training, ranking training, or similar model-improvement use.
An opt-in should describe what content is eligible, what it may be used for, who can authorize it, how long it may be retained, and how to revoke future use.
9. Human review and support access
Human review of Customer Content should be limited. Support access should occur with user/admin permission where practical, or when needed for security, abuse prevention, legal compliance, or emergency response. Users should avoid sending sensitive content to support unless necessary.
10. Builder Vocabulary Engine
QuillCaddie may use vocabulary terms, aliases, likely misrecognitions, and correction metadata to improve transcription accuracy. Raw transcript/source text should remain unchanged. Cleaned transcripts and correction notes are stored separately. Vocabulary processing is part of providing the Service and is not generalized model training.
11. BYOK and customer-managed providers
If BYOK or customer-managed provider routing is enabled, requests may be sent to the provider account configured by the user or workspace administrator. The customer is responsible for that provider’s terms, billing, retention settings, logs, and model-training/model-improvement settings. QuillCaddie is responsible for routing according to configured settings and protecting stored keys using reasonable secret-storage controls.
12. Connectors, MCP, and external AI clients
No external tool should receive raw transcript text, source text, full AI output, files, or Customer Content unless a user/admin explicitly authorizes a connector action, export, sync, context import, or MCP/API tool. External AI clients and third-party tools are governed by their own terms and policies.
13. AI output disclaimer
AI outputs may be wrong. They may omit information, over-summarize, infer incorrectly, mislabel speakers, misunderstand technical terms, or produce unsuitable suggestions. Source material controls over AI outputs. AI outputs are not professional advice. Review all outputs before relying on them, sharing them, publishing them, or sending them to external tools.
14. Deletion and model-training limitations
If Customer Content was never opted into model contribution, deletion follows ordinary retention/deletion flows. If content was explicitly opted into model contribution, revocation generally stops future use but may not remove influence from models already trained or aggregated outputs already created, unless the opt-in terms or applicable law require otherwise.
15. Changes
We may update this Policy as AI features, provider routing, data-use controls, and applicable laws change. Material changes will be communicated by reasonable means.
Contact
Support: support@quillcaddie.com
Privacy requests: privacy@quillcaddie.com
Security reports: security@quillcaddie.com
Legal notices: legal@quillcaddie.com
DMCA notices: dmca@quillcaddie.com
Mail: Bobby Built Ventures LLC d/b/a QuillCaddie, 1500 N Grand St, STE R, Denver, CO 80203