AI Financial Assistant: Privacy and Model Choice

Every finance app is adding AI. Most of them give you a chatbot you can't control; one model, one provider, no transparency about where your data goes. Clarity does it differently. You pick the model. You can bring your own API key. And your financial data never leaves your conversation.

How AI Works in Clarity

Clarity has a built-in AI assistant that can answer questions about your finances using your actual data. It's not a generic chatbot; it has access to 13 financial tools that query your accounts, transactions, holdings, spending, and net worth in real time.

You can ask things like:

• "How much did I spend on food last month?"
• "What's my net worth trend over the last 6 months?"
• "Show me all recurring charges over $50"
• "What's my cash flow this month?"
• "Break down my spending by category"
• "Which subscriptions am I paying for?"

The assistant responds with numbers, charts, and interactive cards, not vague advice. It pulls data directly from your connected accounts using the same database that powers the rest of Clarity. Every answer is grounded in your actual financial data.

The Default: Claude Sonnet

Out of the box, Clarity uses Claude Sonnet from Anthropic. No setup required. No API key needed. It just works.

We chose Claude as the default based on our testing across providers:

• Strong at structured data reasoning. Financial questions require understanding numbers, dates, categories, and relationships between accounts. In our internal testing, Claude handled structured financial queries and tool use well; though other models (GPT-4o, Gemini) are competitive and improving rapidly, which is why we support them too.
• Long context window. When you ask "how has my spending changed this year?", the model needs to hold months of transaction data in context while reasoning about trends. Claude's context window handles this cleanly.
• Reliable tool use. The assistant has 13 tools for querying your financial data. In our experience, Claude is strong at choosing the right tool, formatting parameters correctly, and interpreting results; especially for multi-step queries like "compare my food spending this month to last month" that require two separate tool calls. That said, no model is perfect at this, and you'll occasionally see tool use errors regardless of which model you pick.
• Safety and refusal handling. Claude will summarize your data and identify patterns, but it won't give you specific investment advice or tell you to buy/sell particular assets. This is the behavior we want; the assistant should be a data tool, not a financial advisor.

The default Claude model is included in your Clarity subscription. We pay for the API usage. You don't need to set anything up.

Bring Your Own Key (BYOK)

Not everyone wants to use the default. Maybe you prefer ChatGPT's style. Maybe you want to use a more powerful model like Claude Opus. Maybe you want to try DeepSeek or Gemini. Maybe you just want your API calls going directly to the provider rather than through our key.

Clarity supports Bring Your Own Key for three providers:

Anthropic (Claude)

Add your own Anthropic API key to unlock Claude Opus 4; Anthropic's most capable model. Opus is significantly more expensive than Sonnet ($15 vs $3 per million input tokens), but it's noticeably better at complex financial reasoning: multi-step calculations, tax scenarios, portfolio optimization questions.

With your own Anthropic key, you also get direct billing from Anthropic. You see exactly what you're spending on AI in your Anthropic dashboard, separate from your Clarity subscription.

OpenAI (ChatGPT)

Add your OpenAI API key to use GPT-4o or GPT-4o Mini. GPT-4o is OpenAI's flagship model; fast and capable. GPT-4o Mini is cheaper and faster for simpler questions.

Why you might prefer OpenAI:

• You already have an OpenAI API key and want to consolidate usage under one billing account
• You prefer GPT's conversational style; it tends to be more conversational and less formal than Claude
• Cost optimization; GPT-4o Mini is one of the cheapest capable models available, good for quick balance checks and simple queries

OpenRouter (100+ Models)

OpenRouter is the power user option. It's a unified API that gives you access to every major AI model through one key and one billing account. Add your OpenRouter key and you can switch between:

• Claude Sonnet via OpenRouter; same model, different billing
• Gemini 2.5 Pro from Google; excellent at data analysis, often cheaper than Claude/GPT
• DeepSeek R1; the open-source reasoning model that rivals GPT-4, at a fraction of the cost

OpenRouter also gives you usage tracking across all models in one dashboard. If you're experimenting with different models to find the best fit for financial queries, this is the easiest way to do it.

How Model Selection Works

The model picker is in the chat header. Here's how the selection logic works:

1. You pick a model in the chat UI. The dropdown shows all available models, grouped by provider. Models that require an API key you haven't added are shown but grayed out with an "Add key in Settings" hint.
2. Your choice is saved as your default. Next time you open chat, it starts with your preferred model. You can change it anytime.
3. Fallback to Claude Sonnet. If your selected model requires an API key and the key is missing or invalid, Clarity automatically falls back to the built-in Claude Sonnet. You never get a broken chat experience.

Every message is sent to the model you've selected. There's no batching, no mixing of models mid-conversation. If you switch models mid-conversation, the new model picks up from your message history.

What the AI Can Actually Do

The AI assistant isn't just answering questions from a prompt. It has access to 13 specialized tools that interact with your financial data:

Data Query Tools

• Account Summary; list all accounts with current balances, types, and institutions
• Search Transactions; filter by date range, amount, merchant, or category
• Spending Breakdown; category-level spending totals for any period
• Cash Flow; income vs. expenses with net savings
• Recurring Charges; detected subscriptions and recurring payments
• Net Worth Trend; historical snapshots over time
• Holdings; investment and crypto positions with current value
• Merchant History; all transactions from a specific merchant

Action Tools (Require Your Approval)

Some tools can modify your data. These always ask for explicit confirmation before executing:

• Categorize Transaction; change a transaction's category
• Rename Merchant; set a custom display name for a merchant
• Add Transaction Note; attach a note to any transaction

Export Tools

• Tax Report; generate a Form 8949 CSV with cost basis and wash sale detection
• Bill Calendar; export recurring charges as an .ics calendar file

Generative UI

The AI doesn't just respond with text. It renders interactive charts, metric cards, spending breakdowns, transaction tables, and comparison views directly in the chat. These are the same components used in the rest of Clarity, not generic markdown tables.

When you ask "what's my spending breakdown?", you get a horizontal bar chart with categories and amounts. When you ask "show me my net worth trend", you get a sparkline. When you ask about a specific merchant, you get a transaction list you can scroll through.

Limitations to Be Aware Of

The AI assistant is grounded in your real financial data, but it's still an LLM, and LLMs have known limitations. The numbers it pulls from tools are real, but the model's interpretation of those numbers can occasionally be wrong. It might misattribute a spending increase to the wrong category, summarize a trend incorrectly, or give a confidently worded answer that misses context (like confusing a one-time refund with recurring income).

Treat the assistant as a fast way to query your data and spot patterns, not as an infallible financial analyst. For anything consequential (tax decisions, large purchases, investment changes), verify the numbers in the actual Clarity dashboard or with a professional. The AI is a starting point, not the final word.

Conversations Are Saved

Every conversation is persisted to your account. You can close the tab, come back tomorrow, and pick up where you left off. The chat sidebar shows your conversation history, grouped by time:

• Today
• Yesterday
• This week
• This month
• Older

Conversations are titled automatically based on your first message. You can delete any conversation at any time; it's permanently removed along with all messages.

Privacy: How Your Data Stays Private

This is the section that matters most. When you use AI with financial data, the obvious question is: where does my data go?

What Gets Sent to the AI Model

When you ask a question, the AI model receives your message and the results of any tool calls it makes. For example, if you ask "how much did I spend on food?", the model:

1. Receives your message: "how much did I spend on food?"
2. Decides to call the Spending Breakdown tool with a date range and category filter
3. Receives the tool result: category totals (e.g., "Food & Drink: $847.23")
4. Generates a response with charts and context

The model sees aggregated results from your data, not your raw database. It doesn't get a dump of every transaction. It gets the answer to the specific query it asked for.

What the AI Providers Do With Your Data

This depends on which provider you're using, and whether you're using Clarity's key or your own:

• Anthropic (Claude); via Clarity's key: Anthropic's API terms state that data sent via the API is not used to train models. Your financial data is processed to generate a response and then discarded. Anthropic may retain API logs for abuse prevention (up to 30 days), but your data is not used for training.
• Anthropic; via your key: Same terms apply. You're under your own Anthropic API agreement, which also prohibits training on API data.
• OpenAI (GPT); via your key: OpenAI's API data policy also states that API data is not used for training. Data may be retained for up to 30 days for abuse monitoring, with an option to request zero-retention.
• OpenRouter; via your key: OpenRouter routes your request to the underlying provider. Their privacy depends on the specific model provider's terms, but OpenRouter itself does not train on your data.

What Clarity Stores

On Clarity's side, here's what we store:

• Your conversation messages; stored in our database so you can return to previous conversations. Deleted when you delete the conversation. Deleted when you delete your account.
• Your API keys; encrypted with authenticated encryption before storage. We cannot read your keys at rest. They're decrypted in memory only when making a request to the provider, then discarded. Keys are never logged, never displayed after saving, and never sent to any third party.
• Your model preference, which model you last selected. That's it.

What Clarity Does NOT Do

• We do not send your financial data to AI models unless you're actively chatting
• We do not use your data to train any AI model
• We do not share your conversations with other users
• We do not sell data to third parties
• We do not send your data to a different provider than the one you selected, if you pick OpenAI, the request goes to OpenAI, period

The BYOK Privacy Advantage

Bringing your own API key gives you an additional layer of privacy and control:

• Direct relationship with the provider. Your API calls go from Clarity's server to the provider using your key. The provider sees you as the customer, not Clarity. You're covered by your own API agreement.
• Usage visibility. You can see every API call and its cost in your provider dashboard (Anthropic Console, OpenAI Dashboard, OpenRouter). Full transparency.
• You can revoke access anytime. Delete your key from the provider's dashboard and the connection is severed instantly. No waiting for Clarity to process a cancellation.
• Zero-retention options. Some providers (OpenAI) let you opt into zero-retention policies on your API account. When you use your own key, those policies apply to your Clarity conversations too.

API Key Security

Your API keys are encrypted before they ever touch the database. Here's the technical detail:

• Encryption: Authenticated encryption; the same class of algorithm used by banks, governments, and password managers
• Unique IV per key: Every encryption uses a fresh initialization vector, so even identical keys produce different ciphertext
• Tamper detection: The encryption scheme includes an authentication tag that detects tampering, if anyone modifies the encrypted data, decryption fails
• Encryption key: Stored as an environment variable on the server, never in the database or source code

After you save a key, it's never displayed again. Not in the UI, not in API responses, not in logs. The only time the key is decrypted is in server memory when making a request to the AI provider; then it's immediately discarded.

Rate Limiting

To prevent abuse and runaway costs (especially on the built-in Claude model that Clarity pays for), the chat API is rate-limited per user. The limits are generous enough for normal conversation while preventing automated abuse.

When you use your own API key, the same rate limit applies on Clarity's side. Your provider may have separate rate limits on their end. If you hit the provider's limit, you'll see an error in the chat, not a silent failure.

Why We Built It This Way

Many finance apps that add AI ship a single model integration with limited transparency about data handling. Some are more open than others, but model choice and BYOK are still rare. Here's why we prioritized them:

Models Improve Fast

Six months ago, one model was the best for financial tool use. Now it's a different one. In six more months, it might be something entirely new. By supporting multiple providers and BYOK, we don't have to guess which model will be best — you can switch whenever you want.

Cost Preferences Vary

Some users ask their AI assistant 50 questions a day. Others ask 2 per week. It doesn't make sense to charge everyone the same flat rate for AI, or to bundle a high-cost model into the subscription. The built-in model covers most users. Power users who want more capable models pay the provider directly; at actual API rates, not a marked-up subscription.

Privacy Is Not One-Size-Fits-All

Some users are fine with Clarity's Anthropic key handling their queries. Others want their API calls going directly to a provider they already trust, under their own agreement. BYOK lets privacy-conscious users maintain a direct relationship with their AI provider without Clarity as an intermediary for billing or data processing terms.

Lock-In Is Bad

If we only supported Claude and Anthropic changed their pricing, quality, or terms — you'd be stuck. With multi-provider support, you can switch models in 10 seconds. Your conversations stay in Clarity. Your data stays in Clarity. Only the AI brain changes.

When AI Isn't the Right Tool

Some people reasonably don't want AI involved in their financial data at all — regardless of encryption, privacy architecture, or provider policies. If you're in that camp, Clarity works perfectly fine without the chat feature. Every query the AI can answer is also answerable through the dashboard UI: spending breakdowns, net worth trends, transaction search, recurring charges. The AI is a faster interface to the same data, not a gatekeeper to it.

There are also cases where the AI is less useful than the dashboard: browsing transactions visually, comparing charts side by side, or getting an intuitive feel for your spending patterns over time. The chat is best for specific questions with specific answers. For exploratory analysis, the visual dashboard is usually better.

Getting Started

Use the Default (No Setup)

Just open the chat. Claude Sonnet works immediately for all Clarity users. Ask a question about your finances and get an answer with real data.

Add Your Own API Key

1. Go to Settings → AI & Models
2. Pick a provider: Anthropic, OpenAI, or OpenRouter
3. Paste your API key (it's encrypted immediately and never shown again)
4. Open the chat and select your preferred model from the dropdown

That's it. Your key is encrypted, your model is saved, and every chat message goes through the provider you chose.

Where to Get API Keys

• Anthropic:console.anthropic.com → API Keys → Create Key (starts with sk-ant-)
• OpenAI:platform.openai.com → API Keys → Create New Secret Key (starts with sk-)
• OpenRouter: openrouter.ai → Keys → Create Key (starts with sk-or-)

All three providers offer pay-as-you-go billing. You only pay for what you use. As of early 2026, a typical month of financial queries costs roughly $1-5 on Sonnet, $5-15 on Opus, and under $1 on GPT-4o Mini or DeepSeek — though model pricing changes frequently as providers compete, so check current rates on their pricing pages.

The Bottom Line

AI in personal finance shouldn't be a black box. You should know which model is answering your questions, where your data goes, and have the option to change both. Clarity gives you a capable default (Claude Sonnet, included), the freedom to switch (ChatGPT, Gemini, DeepSeek), and full control over privacy (BYOK).

Your financial data is yours. Your choice of AI model should be too.