Give your agent Solana.

Up in three commands

1. 1 Install

   $ npm i -g @solos-sh/cli

2. 2 Connect a wallet

   $ solos login

3. 3 Connect your agent

   $ solos connect claude   # or: codex | cursor

Restart the client and ask it: what is my SOL balance?

What your agent can do

• wallet · transfer · portfolio

  Balances, addresses, SOL transfers, a portfolio view.

• swap · launch Jupiter, Pump

  Quotes, simulated and executed swaps. Bonding-curve buys and sells.

• lend Kamino

  Reserve rates, positions, deposits and withdrawals.

• liquidity Orca, Raydium, Meteora

  Positions, deposits, withdrawals, opening and closing ranges.

• perp Phoenix

  Positions, equity, collateral, bounded IOC opens and closes.

• market Jupiter, Elfa

  Prices, token metadata, trending tokens, news and research.

The agent sees three tools at start: search, balance, portfolio. It asks for more by name or in plain words, and the matching tools appear. The other 46 cost nothing until they are used.

Same calls, every venue

Jupiter, Pump, Kamino, Orca, Raydium, Meteora and Phoenix each sit behind the same kind of call: a name, a few arguments, a JSON result. The agent learns one way to ask. The layer knows how each venue works, builds the transaction, simulates it, and sends it through one path that signs once. A new venue is a new driver, not a new interface.

You set the limits

• Read and simulate until you say so. Tools that send a transaction do not exist on the server until you connect with --tier execute. Every execute tool has a simulate twin.
• Keys stay on your machine. Client configs carry a profile name, never a secret. A local keypair, Privy, or a pay account signs.
• No rules inside. It does what the agent asks, within the ceiling you set. How much and when is the job of the harness you run it in.