Privacy Policy

1. Introduction

Polydev AI ("we," "our," or "us") is committed to transparency about how we collect, use, and protect your data. This Privacy Policy explains our practices when you use the Polydev AI service ("Service"), including multi-model consultation, connected apps (such as Claude, ChatGPT, Codex, Cursor and VS Code), the local MCP server, Polydev Chat, and the AI model leaderboard.

2. Information We Collect

2.1 Account Information

When you create an account, we collect:

• Email address
• Authentication credentials (encrypted)
• Display name (optional)

2.2 Query and Prompt Data

When you use the Service, we collect:

• Prompts and queries you submit to the multi-model consultation tool
• Model responses returned by each AI provider
• Model selections and preferences — which models you query and any preference signals (e.g., selecting a preferred response)
• Prompt metadata — automatically classified category, programming language, framework, complexity level, and domain

2.3 Leaderboard and Rankings Data

The Service maintains a crowdsourced AI model leaderboard. We collect:

• Pairwise comparison outcomes — which model's response was preferred in each consultation
• Elo rating calculations derived from comparison outcomes
• Prompt dimension classifications — category, language, framework, and complexity labels assigned to each query
• IDE and environment metadata — which IDE or tool you used (e.g., Claude Code, Cursor, Cline)

2.4 Usage Data

We automatically collect:

• API request logs and response times
• Usage counts and billing information
• IP addresses and approximate location
• Device type, browser, and operating system
• Feature usage patterns (which tools you use most)

2.5 API Keys

If you provide your own API keys for third-party services, we store them using industry-standard encryption. We only use these keys to process your requests to the respective services.

2.6 Local AI CLIs and Your ChatGPT Plan

The local MCP server can run AI command-line tools on your own computer, and can use your ChatGPT plan from your own computer, but only for requests made on that computer. We never send requests from connected apps or the hosted MCP server to them. When the local MCP server sends us the answers from your local CLIs or your ChatGPT plan, so they appear in your dashboard and the leaderboard, it sends the answer text only, never your sign-in tokens. We record metadata (model, timestamp, token counts, latency) and keep the content only as described in section 6.

2.7 Connected Apps

You can connect Polydev to an app such as Claude, ChatGPT, Codex, Cursor or VS Code by signing in and approving it. We store the app's name and the access tokens issued to it, and we receive the prompts that app sends to Polydev's tools. You can see and disconnect connected apps at any time under Dashboard → Connected apps.

2.8 Content You Save

Conversations you save in Polydev Chat are kept so you can return to them. Conversation memory, which is on by default and can be turned off in your settings, saves recent exchanges and short notes derived from them so Polydev can give the models context; both are deleted after 30 days.

3. How We Use Your Information

3.1 Service Operation

We use your data to:

• Process your queries and deliver multi-model responses
• Route requests to the appropriate AI providers
• Manage your account and subscriptions
• Enforce rate limits and prevent abuse
• Provide customer support

3.2 Leaderboard and Rankings

We use your query data to:

• Calculate and update Elo ratings for AI models on the public leaderboard
• Generate dimension-specific rankings (by language, framework, category, complexity)
• Publish aggregated model performance statistics
• Improve the accuracy and fairness of our ranking algorithms

To rank models by category, the first 300 characters of each prompt are sent to an OpenAI model that labels its task type, language, framework and complexity. We keep only those labels and a one-way hash of the prompt, not its text.

3.3 Connected Apps

Prompts and responses that arrive through a connected app (section 2.7) are used only to answer that request: they are passed to the model providers you chose and kept only as described in section 6. They are never used to train or evaluate AI models, never included in datasets, and never sold, even in aggregated or anonymized form. Rankings you submit through a connected app may feed the public leaderboard as aggregate counts only.

3.4 AI Training and Model Improvement

For use outside connected apps, we may use aggregated and anonymized data to improve AI systems. This includes:

• Training data. Anonymized prompts, responses, and preference signals may be used to train, fine-tune, or evaluate AI models
• Reinforcement learning. Preference data (which model response was better) may be used for reinforcement learning from human feedback (RLHF) and similar alignment techniques
• Evaluation and benchmarking. Prompts and model outputs may be used to create evaluation datasets and benchmarks for measuring AI model performance
• Routing optimization. Usage patterns help us improve how we select and route queries to different models

We apply anonymization and aggregation before using data for training purposes. We do not publish individual prompts or code in identifiable form.

3.5 Aggregated Data Products

We create and may commercialize aggregated, anonymized datasets derived from Service usage, excluding anything that arrived through a connected app (section 3.3). These may include model performance benchmarks, prompt category distributions, comparative evaluations, and usage trend reports. Aggregated data does not identify individual users or contain identifiable code.

4. Information Sharing

4.1 Third-Party AI Providers

Your prompts are transmitted to third-party AI providers (Anthropic, OpenAI, Google, xAI, and others) to generate responses. Each provider's privacy policy governs their handling of your data. We recommend reviewing their policies.

4.2 Public Leaderboard

Aggregated model rankings, Elo ratings, and performance statistics are published publicly on the leaderboard. This data is derived from usage patterns across all users and does not identify individual users or their specific queries.

4.3 Research and Publications

We may publish research papers, blog posts, or reports based on aggregated data from the Service. These publications will contain only anonymized, aggregated statistics and findings.

4.4 Service Providers

We work with service providers for hosting (Cloudflare, Supabase), payment processing (Stripe), email (Resend), and analytics and error reporting (PostHog, Sentry). These providers have access only to information necessary to perform their functions.

4.5 Legal Requirements

We may disclose your information if required by law, court order, or government regulation, or to protect the rights, property, or safety of Polydev AI, our users, or the public.

5. Data Security

We protect your data with:

• Encryption of data in transit (TLS) and at rest
• Secure API key storage with row-level encryption
• Access controls and authentication for all internal systems
• Regular security assessments

6. Data Retention

Account data is retained while your account is active. You may request deletion at any time.

The text of prompts and responses, including those from connected apps, the local MCP server (answers from your local CLIs or your ChatGPT plan), X search, image generation and conversation memory (with the notes derived from it), is deleted after 30 days. We keep it that long to deliver results, investigate abuse and fix problems. Usage metadata (models, token counts, cost, timings and status) is kept for usage statistics and billing.

Chats you save in Polydev Chat are kept until you delete them or your account.

Anonymized and aggregated data — including leaderboard rankings, prompt category labels, the datasets described in section 3.4, and statistical summaries — is retained indefinitely and survives account deletion, as it cannot be traced back to individual users.

7. Your Rights

Depending on your location, you may have the right to:

• Access your personal data
• Correct inaccurate data
• Request deletion of your account and associated personal data
• Object to or restrict certain processing
• Export your data (data portability)
• Withdraw consent where processing is based on consent

Note: Deletion requests apply to identifiable personal data. Anonymized and aggregated data that has already been incorporated into training datasets, leaderboard calculations, or published research cannot be retroactively removed, as it is no longer linked to your identity.

8. Cookies and Tracking

We use cookies to maintain session state and analyze usage. You can control cookies through your browser settings, but some features may not function properly with cookies disabled.

9. Children's Privacy

The Service is not intended for children under 13. We do not knowingly collect personal information from children. If you believe we have collected information from a child, contact us immediately.

10. International Data Transfers

Your data may be processed in the United States and other countries where our service providers operate. We ensure appropriate safeguards are in place for international transfers.

11. Changes to This Policy

We may update this Privacy Policy periodically. Material changes will be communicated via email or through the Service. Continued use after changes constitutes acceptance.

12. Contact Us

For questions about this Privacy Policy, to exercise your data rights, or to request information about our data practices, contact us at support@polydev.ai or through our dashboard.