Privacy Policy

2.1 Information You Provide

• Account Information: Name, email address, and authentication credentials when you create an account (managed through Clerk, our authentication provider)
• AI Provider API Keys (optional): If you choose to bring your own keys, the key you enter is encrypted at rest and stored against your account. It is never displayed back to you and is never sent to your browser. We use it only server-side to make the AI calls you initiate. We do not use your key for background work: scheduled jobs such as the weekly digest run on TradeHarbor's own provider accounts, so nothing is billed to you while you are away. You can remove a key at any time from Settings, and revoke it from the provider's own console.
• Profile Data: Any optional profile information you provide, such as display name or preferences
• Communications: Messages, feedback, or support requests you send to us
• Watchlist and Portfolio Data: Tickers you follow, watchlist configurations, portfolio allocations, and trading preferences you set within the Platform
• Paper Trading Activity: Paper trades you accept or reject, entry and exit prices, hold durations, realized and unrealized P&L, and any outcome labels generated by the Platform
• Trade Journal Entries: Free-text notes, tags, sentiment markers, and observations you write about your trades, as well as AI-generated post-mortems attached to your entries
• Assistant Conversations: Messages you send to the TradeHarbor Assistant (the in-app AI chat). Conversation history is held in-memory for the current browser session only and is not persisted in our database.

2.2 Information Collected Automatically

• Usage Data: Pages visited, features used, recommendations viewed, and interactions with the Platform
• Device Information: Browser type, operating system, device type, and screen resolution
• Log Data: IP address, access times, referring URLs, and error logs
• Cookies and Similar Technologies: As described in our Cookie Policy

2.3 Information from Third Parties

• Authentication Provider: We receive basic profile information from Clerk when you sign in using third-party authentication (e.g., Google, GitHub)

5.1 Tool Calling

When you ask the assistant a question that requires knowing your data, it may invoke server-side "tools" that fetch a narrow, user-scoped slice of your data (for example, your open positions, a specific recommendation, or a recent journal entry). Every tool call is authenticated against your account and scoped to data you own. The assistant cannot read data belonging to other users.

The retrieved data is then included in the prompt sent to the third-party large language model (LLM) that generates the assistant's response. Which LLM provider is used depends first on whether you have stored your own provider keys — if you have, only providers you hold a key for are used — and otherwise on your configured preference and our failover chain (see Section 6 below).

5.2 Semantic Retrieval (Embeddings)

To let the assistant find the most relevant past trades and journal entries for a given question, we generate numerical "embeddings" of your Trade Journal Entries and the AI-generated reasoning attached to your Recommendations. These embeddings are produced by OpenAI's text-embedding-3-small model and stored in our database alongside the original row, with your user ID attached. Embeddings are used only for user-scoped semantic search: when you ask the assistant a question, we embed the question and compare it against your own embeddings to surface your most similar past entries.

Embeddings are generated by a nightly background job and only for content that belongs to a real user account. Content generated by our internal data-collection pipelines (for example, synthetic recommendations used to train our ML model) is never embedded and is never returned in any user's retrieval results.

5.3 Scope Guardrail

The TradeHarbor Assistant is designed as an analysis and explanation tool, not a recommendation generator. It will not invent new trade ideas in chat. Questions that ask for new recommendations are redirected to the dedicated scan interface, which is subject to the same disclaimers set out in our Investment Disclaimer.

5.4 Your Controls

You have several controls over the assistant's use of your data:

• You are not required to use the assistant. If you do not open it or send it messages, no tool calls are made and no queries are embedded on your behalf.
• You can delete all embeddings generated from your content at any time via the Settings page. This removes the semantic index over your data; the underlying journal and recommendation rows remain until you delete them separately.
• You can delete your account as described in Section 8, which removes both your source content and any embeddings derived from it.
• We maintain administrative kill switches that allow us to disable tool calling or semantic retrieval across the Platform in the event of a technical or privacy incident.

6.1 Service Providers

• Clerk: Authentication and user management.
• Large Language Model Providers: We use third-party LLMs to generate trade research, the TradeHarbor Assistant responses, sell signals, post-mortems, and other AI features. Scans and the Trading Room use the models you switch on in Settings: on your own provider key where you hold one, and otherwise on TradeHarbor's provider accounts as your plan allows. The TradeHarbor Assistant, chart readings, sell signals, post-mortems, the reasoning check and assignment reviews always run on TradeHarbor's own provider accounts, on a model we choose for your plan, never on a key you supplied. When you interact with the TradeHarbor Assistant in a way that requires your personal data (for example, asking about your portfolio), the specific slice of your data returned by the relevant server-side tool is included in the prompt sent to the LLM provider so that the model can produce a grounded answer. Where a call is made on TradeHarbor's own provider accounts, it goes only to these providers, and we rely on each one's published data-handling terms:
  • Anthropic (Claude): processes and stores data in the United States; does not use TradeHarbor API traffic to train its models by default.
  • OpenAI (GPT): processes and stores data in the United States; does not use TradeHarbor API traffic to train its models by default.
  • Google (Gemini, on its paid API tier): processes and stores data on Google's infrastructure, in the United States and other countries; does not use TradeHarbor API traffic on the paid tier to train its models.
  • Groq (GPT-OSS): processes data in the United States; does not use TradeHarbor API traffic to train its models.
  • DeepSeek: processes and stores data in the People's Republic of China, and its API terms don't rule out using the data it receives to improve its models, so assume it may. It runs a scan as the Free plan's low-cost model, and a scan or a Trading Room seat when you switch DeepSeek on in Settings.
  A scan sends market data and the scan's settings, including a risk level (conservative, moderate or aggressive) taken from your sizing rule. It never sends your positions, balance, journal or account details. The Trading Room sends what you type, and your paper positions and balance when you ask for them with /positions, /position or /account. Other providers, such as xAI, Moonshot and Zhipu, are available for scans and the Trading Room only when you bring your own key. Where a call is made with a key you supplied (scans and the Trading Room), it reaches that provider on your account and is governed by your own agreement with them, including whatever data-retention and training settings apply to that account — which are yours to control, not ours.
• OpenAI (Embeddings): In addition to its LLM role above, OpenAI's embedding API is used to generate semantic embeddings of your Trade Journal Entries and Recommendation reasoning as described in Section 5.2. Embedding requests contain the text of your entries; OpenAI does not use embedding API inputs to train its models.
• Market Data Providers: Yahoo Finance, Alpaca, and Tradier for financial data. No personal data is shared with these providers.
• ML Prediction Service: Our own machine learning service (hosted on Railway) receives numerical feature vectors and recommendation identifiers in order to return P&L probability predictions. No free-text user content is sent to this service.
• Hosting Provider: For infrastructure and data storage.

6.2 Legal Requirements

We may disclose your information if required by law, court order, or governmental authority, or if we believe disclosure is necessary to protect our rights, your safety, or the safety of others.

6.3 Business Transfers

In the event of a merger, acquisition, or sale of assets, your personal data may be transferred as part of the transaction. We will notify you of any such change.

9.1 GDPR Rights (EEA/UK Residents)

• Access: Request a copy of the personal data we hold about you
• Rectification: Request correction of inaccurate or incomplete data
• Erasure: Request deletion of your personal data ("right to be forgotten")
• Restriction: Request restriction of processing in certain circumstances
• Portability: Request your data in a structured, machine-readable format
• Objection: Object to processing based on legitimate interests
• Withdraw Consent: Withdraw consent at any time where processing is based on consent
• Complaint: Lodge a complaint with your local Data Protection Authority

9.2 CCPA Rights (California Residents)

• Right to know what personal information is collected, used, and shared
• Right to delete personal information
• Right to opt out of the sale of personal information (we do not sell personal data)
• Right to non-discrimination for exercising your privacy rights

9.3 Assistant Embedding Deletion

Independent of the rights above, any logged-in user may delete all semantic embeddings generated from their Trade Journal Entries and Recommendations at any time. This wipes the retrieval index used by the TradeHarbor Assistant's semantic search feature. The underlying entries remain unless you delete them separately.