Introduction
Solrouter is private AI chat. Prompts are sealed on your device and opened only inside an Intel TDX enclave on Phala that you can verify.
Solrouter is private AI chat. On the sealed path your prompt is encrypted on your device, relayed as ciphertext by Solrouter's backend, and opened only inside an Intel TDX enclave on Phala. Each sealed call writes a receipt to Solana. Use it in the web chat at solrouter.com, Claude Code, Cursor, Claude Desktop, or your own app.
In words
- A typical AI API reads your prompt as plain text on its own server.
- With Solrouter, your device seals the prompt first. The backend relays ciphertext and has no key to open it.
- Only the enclave holds the key. It calls the model on a GPU node, then seals the reply so only your device can read it.
What is sealed
| Surface | Sealed |
|---|---|
TypeScript SDK, client.chat() | Yes |
MCP server, encrypted_chat and the model step of the research tools | Yes |
| Web chat at solrouter.com, signed in or guest | Yes |
| Agent mode in the web chat, and its Notion and GitHub connectors | No |
MCP web search and market lookups, agent_run | No |
Sealed means X25519 key agreement per request plus Arcium RescueCipher on your device, decryption only inside the enclave, and a Light Protocol compressed attestation on Solana under program ATMRatMtsKX4bHax7U4FRdhbE4mjU4NKpDZGqZqAhBKb. The full breakdown is on What is private here.
Open-weight models only
Solrouter serves GPT-OSS 20B, Qwen 3.8 27B, and Qwen3 Coder 30B. No prompt goes to OpenAI, Anthropic, or Google. The GPU node that runs the model sees the prompt while it works on it, and Solrouter does not control that hardware.
Start here
Integrations
Claude Code, Cursor, Claude Desktop, the TypeScript SDK, Notion, and GitHub.
Pricing
Pro $9, Builder $49, Scale $99 a month in USDC. 5 free messages to start.
Models
Three open-weight models, 8K context and 4K output.
Verify the enclave
Check the enclave's key against its Intel TDX quote in your browser.
New to the tech? Start with What is a TEE?.