DCP

Your keys never leave the device. DCP signs locally and uploads nothing, which matters if you don't want a provider holding the thing that controls your money. The vault can also store non-wallet data like identity fields and API credentials, each private until you approve a request for it.

Every agent you connect gets its own permissions. One might read a single credential. Another might hold a small wallet budget, and a research agent might only read approved notes. You decide what each one can use, spend, and read, and the dashboard keeps all of it in one place.

Set a per-transaction and daily cap, and DCP splits actions into two lanes. Anything inside your rules runs automatically; anything over the limit or out of the ordinary asks for approval. An agent can't drain the wallet because the cap holds regardless of what it tries.

Connect Telegram and sensitive requests reach you anywhere. You get a message like "your assistant wants to send $50," then tap Approve or Deny. Not ideal for every purchase. Each approval is tied to the specific amount and destination you see, so it can't be quietly swapped for something else.

Connecting an agent takes a single line, and it shows up in your dashboard with its own rules about a minute later. The docs cover stdio setup for Claude Desktop, Cursor, and VS Code, plus a streamable HTTP mode for remote or URL-based clients. No plugins and no keys handed to the agent.

Every request, check, and result is logged, so you can see what an agent did and when. If one starts behaving oddly, one click cuts off its access while the rest keep working. That's the part most people don't think about until something goes wrong.

DCP is built around Solana, so a Solana agent wallet here means an agent that can check balances, read transaction history, transfer SOL or SPL tokens, swap through Jupiter, sign transactions, and sign x402 payment payloads. Wallet reads need no approval; anything that moves value goes through budget and approval first.