AI Providers

getbased supports six AI backends for PDF import, chat, and dashboard AI features. You can switch between them at any time in Settings → AI.

Which Features Need an AI Provider?

Feature	Requires AI?
PDF import	Yes
AI chat panel	Yes
Focus card (dashboard insight)	Yes
Health status dots on context cards	Yes
AI-generated card tips	Yes
Web search in chat	Yes (OpenRouter, PPQ, Venice)
Charts, tables, trend alerts	No
Manual entry	No
JSON export / import	No
Correlations, compare dates	No
Marker glossary	No

All non-AI features work fully without any provider configured.

The Six Providers

PPQ (Anonymous, crypto)

A pay-per-query AI aggregator with 300+ models. No subscription, no KYC. Top up with Bitcoin, Lightning, Monero, Litecoin, Aqua, or Bitrefill gift cards — directly in the app. Balance displayed in settings. Supports web search in chat.

Setup:

1. In Settings, select PPQ
2. Click Create Account — instant, no signup
3. Save your API key — PPQ accounts are anonymous with no recovery
4. Top up via the Top Up button (Lightning, Bitcoin, Monero, Litecoin, Aqua)
5. Choose a model from the curated dropdown

Or paste an existing API key from ppq.ai.

In-app topup

After creating an account, click Top Up to fund it with crypto. Your balance is shown with color coding — the app walks you through the payment flow with QR codes and payment polling.

Routstr (Decentralized Bitcoin AI)

Decentralized AI using Bitcoin micropayments. No account, no sign-up, no subscription. getbased has a built-in Cashu eCash wallet — fund it with Lightning, then connect to any Routstr node discovered via Nostr relays. Your prompts go directly from your browser to the node you choose.

Setup:

1. In Settings, select Routstr
2. Fund your wallet — click Deposit and pay the Lightning QR code (or paste a Cashu token)
3. Pick a node — the app discovers online nodes via Nostr. Click Connect on any node
4. Deposit sats — choose how much to deposit to the node. You get a session key
5. Start chatting — your wallet balance and node session balance are shown separately

Wallet features:

• Seed phrase — 12-word BIP-39 mnemonic generated on first deposit. Write it down — it's the only way to recover your wallet
• Lightning withdraw — send sats to a Lightning address or pay a BOLT11 invoice
• Cashu token send/receive — withdraw as a shareable Cashu token, or deposit one from an external wallet
• Node withdraw — pull remaining sats back from a node into your wallet
• Mint selection — change the Cashu mint (the app auto-switches when a node requires a specific mint)
• Recovery — if a deposit or withdrawal fails mid-operation, a recovery banner appears with your token

Wallet vs Node balance

Your wallet balance is sats you hold locally (Cashu proofs in your browser). Your node session balance is sats deposited to a specific Routstr node for API calls. You can move sats between them freely.

Back up your seed phrase

The seed phrase is shown once when you first deposit. You can view it again in the wallet menu (Seed & Restore). Without it, your wallet cannot be recovered on another device.

OpenRouter (Most models, card/USDC)

A model marketplace with 200+ models — Claude, GPT, Gemini, DeepSeek, Grok, Qwen, and more. Pay with card or USDC. Balance displayed in settings. Supports web search in chat.

Setup — OAuth (easiest):

1. In Settings, select OpenRouter
2. Click Connect with OpenRouter
3. Authorize getbased on the OpenRouter site
4. You're connected — no API key needed

Setup — API key:

1. Get an API key at openrouter.ai
2. In Settings, select OpenRouter
3. Paste your API key
4. Choose a model from the curated dropdown, or type any model ID into the custom input field

One-click connect

The OAuth button also appears in the chat setup guide when no provider is configured.

Venice AI (Best for privacy)

A privacy-focused cloud provider where your conversations and data are not stored or logged on their servers. Venice also proxies access to GPT, Grok, and DeepSeek models. Supports web search in chat with any model.

Setup:

1. Get an API key at venice.ai
2. In Settings → AI Provider, select Venice
3. Paste your API key
4. Choose a model from the dropdown

End-to-End Encryption

Venice offers E2EE models where your prompts are encrypted in your browser and only decrypted inside a verified Intel TDX Trusted Execution Environment — not even Venice can read them. Enable the End-to-End Encryption toggle in Venice settings to switch to E2EE models. The TEE attestation is verified client-side on every session (nonce binding, signing key binding, debug mode rejection). A 🔒✓ (green checkmark) in the chat header confirms verification passed. Web search and image attachments are disabled in E2EE mode.

Custom API (Any endpoint)

Connect to any OpenAI-compatible API endpoint with your own base URL and API key. Works with OpenAI, Mistral, Groq, Together, xAI, OpenCode, Fireworks, Deepinfra, vLLM, LiteLLM, and any other service that implements the /v1/chat/completions standard.

Setup:

1. In Settings, select Custom
2. Enter the Base URL of your endpoint (e.g. https://api.openai.com/v1)
3. Enter your API key
4. Click Save & Validate — the app checks your key and fetches the model list
5. Pick a model from the dropdown, or type any model ID into the manual input

If the endpoint doesn't expose a /v1/models listing (e.g. OpenCode Go), validation still succeeds and you can enter the model ID manually.

When to use Custom vs other providers

Use Custom when you have a direct API key for a service that isn't one of the built-in providers. If the service is available through OpenRouter or PPQ, those providers are easier — they handle model discovery and show pricing automatically.

Local endpoints work too

If your Custom API endpoint is on your local network (localhost, 192.168.*), requests go directly without the CORS proxy. For truly local-only AI, use the dedicated Local provider instead — it has Model Advisor and CORS setup guidance built in.

Local AI (Fully local)

Run a language model entirely on your own machine. Nothing is sent over the network — not even for PDF import. Local AI connects via the standard OpenAI-compatible API (/v1/chat/completions), which is supported by all major local servers:

• Ollama — easiest setup, pull models from the command line
• LM Studio — GUI-based, drag-and-drop model loading
• Jan — open-source desktop app
• llama.cpp, LocalAI, and others

Setup:

1. Install and start any local AI server
2. Load a model (e.g., in Ollama: ollama pull llama3.2)
3. In Settings → AI Provider, select Local
4. Enter your server URL (default: http://localhost:11434)
5. Click Test — the app auto-discovers available models
6. Add an API key if your server requires one (most don't)

Ollama Cloud models are not local

Ollama supports :cloud models that run on Ollama's servers, not your machine. These appear in the model dropdown if you've pulled them, but your data leaves your device when using them. If privacy is why you chose Local AI, stick with locally-running models. The Model Advisor marks cloud models with a ☁ badge so you can tell them apart.

Local AI also handles PII stripping

When enabled in Settings → Privacy, your local server is used to intelligently strip personal information from PDFs before analysis. See PII Obfuscation for details.

Cross-origin (CORS) access

Local AI servers block requests from web pages by default. The app detects this and shows OS-specific instructions, but here's the quick reference:

Ollama:

• Linux: OLLAMA_ORIGINS=* ollama serve
• macOS: launchctl setenv OLLAMA_ORIGINS "*" then restart Ollama.app
• Windows: Add OLLAMA_ORIGINS = * as a system environment variable, then restart Ollama

LM Studio: Settings → Enable CORS

Jan: Settings → Advanced → Enable CORS

HTTPS limits Local AI to localhost

The hosted app at app.getbased.health is served over HTTPS. Browsers block HTTPS pages from making requests to plain HTTP servers on your LAN (mixed content). This means Local AI must run on the same machine — only localhost / 127.0.0.1 will work. If you need to reach a server on another device, use the local dev server (node dev-server.js) which runs over HTTP.

Model Advisor

When connected to Ollama, the app detects your GPU and shows a Model Advisor panel below the model dropdown. Each installed model gets a fitness rating for lab analysis (★ Recommended, Capable, Underpowered, or Inadequate) and a VRAM fit check. If none of your models are recommended, it suggests the best one to pull for your hardware. For remote Ollama servers, enter the server's VRAM manually to get accurate recommendations.

Use a capable model

Models under 14B parameters struggle with accurate marker extraction from complex lab PDFs. The Model Advisor in Settings will tell you exactly which of your installed models are suitable — look for the ★ Recommended badge. When in doubt, ollama pull qwen2.5:14b is the best value for reliable local lab parsing.

Recommended Models

All providers show a tiered model dropdown with two groups:

• ★ Recommended — the latest, most capable models for lab interpretation (sorted first)
• Other — all remaining available models

Recommended models are chosen for accuracy with medical/scientific data. You can use any model, but recommended ones produce the most reliable results.

Model Consistency

Use the same model across imports

When you import lab PDFs, the AI generates marker keys (like biochemistry.glucose) to map results. Different models may generate slightly different keys for the same marker. Switching models between imports can cause the same marker to appear as two separate entries.

For best results, pick a model and stick with it for all your imports. If you do switch, getbased runs a pre-flight check before each import and warns you if your model has changed since the last import.

How Much Does It Cost?

AI providers charge you based on how much text you send and receive — but the amounts are tiny. getbased shows the exact cost of every interaction in the chat panel, so you always know what you're spending.

Here's what real usage costs with the recommended models:

Model	Provider	Import a lab PDF	Chat message	First-time setup*	Ongoing month**
Claude Sonnet 4.6	OpenRouter / Routstr / PPQ	~$0.04	~$0.02	~$1.00	~$0.50
GPT 5.4	OpenRouter / Venice / Routstr / PPQ	~$0.03	~$0.02	~$0.80	~$0.45
Gemini 3.1 Pro	OpenRouter / Venice / Routstr / PPQ	~$0.03	~$0.01	~$0.60	~$0.35
Grok 4	OpenRouter / Venice / Routstr / PPQ	~$0.01	~$0.005	~$0.25	~$0.15
Any model	Custom API (direct key)	Varies	Varies	Varies	Varies
Any local model	Local AI (Ollama, LM Studio)	Free	Free	Free	Free

* First-time setup: importing your first labs + setting up your profile through chat (health goals, context cards, interpretive lens) — typically 3–5 imports and 30+ chat messages.

** Ongoing month: 2–3 lab imports, 20–30 chat messages, dashboard AI features (focus card, health dots). Heavy users who chat daily may spend 2–3x more.

No surprises

Every AI response in getbased shows its cost right below the message. You can see exactly what each import and chat message costs as you use the app. Most users spend well under $1/month. If your credits run out, getbased shows a clear message with a link to add more.

Free option

Run a local model with Ollama, LM Studio, or Jan and pay nothing. You'll need 8GB+ VRAM (or a Mac with 16GB+ unified memory) for capable models. The Model Advisor in Settings shows exactly what fits your hardware.

Switching Providers

You can switch providers at any time in Settings without losing any data. Your API keys are stored locally in your browser and are never sent to getbased servers.