thinyai

A beautiful terminal AI agent — interactive chat, tools, cross-session memory on Walrus, and grounded Sui execution. The command is thiny.

 _____ _     _                _    ___
|_   _| |__ (_)_ __  _   _   / \  |_ _|
  | | | '_ \| | '_ \| | | | / _ \  | |
  | | | | | | | | | | |_| |/ ___ \ | |
  |_| |_| |_|_|_| |_|\__, /_/   \_\___|
                     |___/

Install

bun add -g thinyai      # or:  npm i -g thinyai   /   pnpm add -g thinyai

This installs one command: thiny. No .env, no config files to hand-edit — the first run sets you up.

thiny                   # first launch runs setup, then starts chatting

Commands

| Command | What it does | |---|---| | thiny | Start the interactive agent. Runs first-time setup if needed. | | thiny init | (Re)run base setup — pick a model + agent name + API key. | | thiny sui init | Add Sui on-chain capabilities — pick a network + wallet. | | thiny update | Update to the latest published version (auto-detects bun/npm/pnpm). | | thiny help | Show all commands. | | thiny --version | Print the version. |

Everything is saved to ~/.thiny/config.json (chmod 0600 — it holds your API key and any Sui key). No .env required.

First run / thiny init

The first time you run thiny (or any time you run thiny init), an arrow-key wizard asks:

1. Agent name — what the assistant calls itself (default ThinyAI).

2. Model — pick from the list, or Custom for any OpenAI-compatible endpoint:

   | Choice | Notes | |---|---| | OpenAI · gpt-4o-mini | fast, cheap | | OpenAI · gpt-4o | | | Anthropic · claude-haiku-4-5 | | | Anthropic · claude-sonnet-4-6 | | | Ollama | local, no key (http://localhost:11434/v1) | | Custom | enter model id + base URL + key — works with Groq, Together, OpenRouter, LM Studio, vLLM, Mimo, Azure, … |

3. API key — pasted securely (masked); skipped for keyless setups like Ollama.

That's it — thiny starts.

Sui setup / thiny sui init

Adds grounded, capped on-chain execution. If you haven't run base setup yet, this runs it first.

1. Network — Testnet (recommended) or Mainnet. You can change this later by re-running thiny sui init.

2. Wallet — choose one:

   | Option | What happens | |---|---| | Paste an existing private key | Enter a suiprivkey… key (masked). | | Generate a new key pair locally | A fresh Ed25519 key is created and stored in your config. | | Agent wallet (Rill) | Generates a signer key and stores your per-user Rill MCP URL for grounded execution. |

After setup it prints your address — you must fund it before sending transactions:

⚠ Fund this address (testnet) before sending transactions
  0xabc…
  Faucet: https://faucet.sui.io   (or `sui client faucet`)

Mainnet is gated: the signer refuses to sign on mainnet unless explicitly enabled, and per-tx / budget / expiry caps are enforced on-chain by the agent-wallet contract — the local policy is only a soft UX layer.

In-chat commands

While chatting, type a slash command:

| Command | Description | |---|---| | /new | Start a fresh session (long-term memory carries over). | | /tools | List available tools. | | /skills | List available skills. | | /session | Show the current session id. | | /stats | Session totals (turns, tokens, tool calls). | | /verify <blobId> | Re-fetch and replay a stored Walrus audit trail. | | /clear | Clear the screen. | | /help | Show these commands. |

Press Ctrl-D (or Ctrl-C) to exit.

Memory (Walrus)

Thiny remembers across sessions. Durable facts about you are stored on Walrus (content-addressed, verifiable, portable — not locked to one machine) and auto-injected at the start of each conversation. When a fact is saved you'll see a compact, verifiable line:

✓ memory saved on Walrus · https://walruscan.com/testnet/blob/…

Writes are non-blocking — you can keep chatting while they upload. Set MEMWAL_* env vars to use MemWal (semantic memory) instead.

Output

Responses render as markdown in the terminal — bold, italics, inline code, lists, headings, blockquotes, fenced code blocks, and clickable [links](url). Model reasoning (<think>…</think>) streams dimmed.

Config reference

~/.thiny/config.json:

{
  "agentName": "ThinyAI",
  "userId": "default",
  "model": "openai:gpt-4o-mini",   // or "anthropic:…", or a bare id with baseUrl
  "baseUrl": "https://…/v1",        // optional — for OpenAI-compatible endpoints
  "apiKey": "sk-…",
  "sui": {                          // present after `thiny sui init`
    "network": "testnet",
    "address": "0x…",
    "wallet": { "type": "generated", "secretKey": "suiprivkey…" },
    "rillMcpUrl": "https://…"       // optional — Rill agent wallet
  }
}

Environment variables override the config when set (handy for CI/dev): THINY_MODEL, THINY_OPENAI_API_KEY, THINY_ANTHROPIC_API_KEY, THINY_OPENAI_BASE_URL, THINY_PERSONA_NAME, THINY_USER_ID, SUI_NETWORK, SUI_SECRET_KEY, MCP_URL.

Troubleshooting

• (model returned empty response) / API error — wrong model id, base URL, or key. Re-run thiny init, or edit ~/.thiny/config.json. The CLI prints the underlying provider error.
• Transactions fail — make sure the address from thiny sui init is funded.
• Logs — written to ~/.thiny/cli.log (never to the chat). Tail with tail -f ~/.thiny/cli.log.

Part of the Thiny framework — a tiny, hexagonal agent kernel for Web2 + Web3.