Ask anything (AI agent)
splice ask answers questions about Robinhood Chain, DEX markets, Robinhood Stock Tokens, oracle prices, wallets, the web and GitHub. The AI model decides which of Splice's read-only live data tools to call, Splice runs them against the real providers, and the model answers from the results.
splice ask "Top 5 gainers on Robinhood Chain in the last 24 hours"
splice ask "What are NVDA and TSLA trading at on Robinhood Chain?"
splice ask "Which tokens are losing volume right now?"
splice ask "Inspect wallet 0xa4b000000000000000000073657175656e636572"
splice chat
What you see
- Each tool call as it happens, with its arguments, status and provider, for example
✓ market_movers kind=gainers → LIVE coingecko. - The answer (Markdown rendered for the terminal).
- A footer: number of tool calls, the providers used, the model and the cost reported by the AI provider, for example
2 tool calls · sources: coingecko, dexscreener · model openai/gpt-4o-mini · cost 0.002128 USD.
--json prints the question, answer, tool calls (name, arguments, status, source), model and cost.
How it stays honest
- The model is instructed to call tools for every price, market, token, wallet, stock, oracle, web or GitHub fact and never to answer them from memory.
- Tool results carry status (
LIVE,CACHED,UNAVAILABLE,ERROR) and provenance; when a source is unavailable the answer says so. - Only read-only tools are available. The agent cannot send transactions, write files or call another model.
- Tool results are trimmed before they are sent to the model (long lists, ABIs, traces).
- Market data is not financial advice; the agent does not tell you what to buy or sell.
Tools
The same tools splice mcp --data exposes, except ai_generate and ai_models:
| Area | Tools |
|---|---|
| Markets | market_trending, market_new_pools, market_top_pools, market_movers, market_get_price, market_token, market_pairs, market_token_price, market_ohlcv, market_search |
| Stock tokens | stock_list, stock_quote, stock_movers |
| Oracle | oracle_price, oracle_candles, oracle_feeds |
| Tokens (Codex) | tokens_rank, tokens_search, token_details, token_trades, token_chart |
| Global | global_overview, global_coins, global_trending, global_stocks |
| Perps (Lighter) | perps_markets, perps_funding |
| Companies, news, macro | stock_profile, stock_news, market_news, earnings_calendar, us_market_status, macro_overview, macro_series |
| DeFi (DefiLlama) | defi_overview, defi_protocols, defi_dexes, defi_fees, defi_yields, defi_stablecoins, defi_token_price, defi_price_chart |
| Chain and wallets | chain_info, onchain_get_block, onchain_get_transaction, onchain_get_balance, onchain_get_token, onchain_get_contract, onchain_get_transfers, onchain_get_logs, wallet_inspect, wallet_get_portfolio, security_get_token, security_get_address |
| Web and GitHub | web_search, web_extract, web_answer, web_map, web_similar, github_repository, github_search_repositories, github_contents, github_commits, github_releases, github_raw |
| Providers | providers_status |
Model and cost
| Variable | Purpose |
|---|---|
OPENROUTER_API_KEY | Required (or GEMINI_API_KEY with --provider gemini). |
AI_ASK_MODEL | Model for ask/chat only, for example a stronger model than the default. |
AI_DEFAULT_MODEL | Used when AI_ASK_MODEL is not set. |
| — | Without either, openai/gpt-4o-mini on OpenRouter (gemini-2.5-flash on Gemini). |
--model <id> overrides all of them for one question. A question typically takes one to three tool calls and two to four model calls; the cost is printed with every answer. Some tools use billed provider credits as well (web search and extraction).