AI Market Sentiment
Classify a short text's sentiment toward a specific asset — a crypto coin OR a US stock ticker — with live market context (price, trend, crypto Fear & Greed, equity/cashtag identity) gathered automatically so the model resolves market-native rhetoric — 'inverse BTC', contrarian/fade language, cashtag proxies — against actual market direction. Subjects resolve via CoinGecko (crypto) with a Finnhub equity fallback for tickers like AAPL/NVDA. Runs on Caddie's task-graph runtime; the action is a signed proxy. Single LLM call, fixed structured output. Ideal for: tweet/headline-triggered trading workflows, sentiment-based alerts, contrarian signal detection. Drop-in replacement for an AI Prompt node doing sentiment classification — the failure mode it fixes is the model never seeing market state.
Catalog action Utility
Classify a short text's sentiment toward a specific asset — a crypto coin OR a US stock ticker — with live market context (price, trend, crypto Fear & Greed, equity/cashtag identity) gathered automatically so the model resolves market-native rhetoric — 'inverse BTC', contrarian/fade language, cashtag proxies — against actual market direction. Subjects resolve via CoinGecko (crypto) with a Finnhub equity fallback for tickers like AAPL/NVDA. Runs on Caddie's task-graph runtime; the action is a signed proxy. Single LLM call, fixed structured output. Ideal for: tweet/headline-triggered trading workflows, sentiment-based alerts, contrarian signal detection. Drop-in replacement for an AI Prompt node doing sentiment classification — the failure mode it fixes is the model never seeing market state.
At a Glance
| Field | Value |
|---|---|
| Action ID | ai-market-sentiment |
| Category | Utility |
| Connector | Not required |
| Requires gas | No |
| Funds movement | None declared |
| Tags | ai, llm, sentiment, openrouter, utility, market, crypto, transform, text |
Payload Schema
| Field | Type | Required | Description |
|---|---|---|---|
text | string | Yes | The content to classify (tweet, headline, news snippet, chat message). Pass in trigger output (e.g., {{root.result.text}} for an x-new-tweet trigger). |
subject | string | Yes | The asset(s) the sentiment is toward — a crypto coin (name, symbol, CoinGecko id, or contract address, e.g. 'BTC', 'bitcoin', '0xabc...') OR a US stock ticker (e.g. 'AAPL', 'MARA', 'NVDA'). Pass a COMMA-SEPARATED list (e.g. 'BTC,ETH,SOL' or 'AAPL, NVDA', up to 5) to classify each independently — results come back in 'subjects[]'. Each is resolved to a CoinGecko id for crypto; if not a coin, it falls back to a Finnhub equity quote. On a full miss, classification uses bare-text reasoning. |
author | string | No | Optional author handle/byline (e.g., '@VitalikButerin'). Surface hint to the model only — v1 does not maintain a per-handle reputation map. |
includeMarketContext | boolean | No | Master switch for the deterministic market-data gather phase. When false, the action runs as a bare-text sentiment classifier with no market fetches. Default: true. |
additionalContext | string | No | Extra context from prior workflow steps to pass to the model (same semantics as AI Prompt's 'context'). Stringify JSON if needed. |
customGuidance | string | No | Optional domain guidance appended to the system prompt (e.g., 'This account is known for sarcasm', 'Treat $MARA as a Bitcoin-correlated mining proxy'). Use sparingly — the prompt already covers crypto rhetoric (inverse, contrarian, cashtag proxies). TRUST NOTE: this is operator-authored guidance and is injected OUTSIDE the prompt-injection guard rail (unlike 'text'/'author'/'additionalContext'), so set it from workflow config — never wire untrusted user/tweet content into it. |
model | string | No | The OpenRouter model to use (default: 'anthropic/claude-haiku-4.5'). Any model id from the OpenRouter catalog is accepted. When unset, Caddie runs its configured classifier model. |
temperature | number | No | Creativity/randomness for the classification call. Default 0.2 — sentiment should be deterministic, not creative. |
maxTokens | number | No | Maximum tokens in the response (default: 1024, max: 16384). |
Result Schema
| Field | Type | Required | Description |
|---|---|---|---|
sentiment | string | Yes | Classification toward the subject asset. |
confidence | number | Yes | Model-reported confidence in the classification (0.0-1.0). |
reasoning | string | Yes | Short explanation of how the model resolved the text — including market direction usage when relevant. |
marketContext | object | Yes | Audit trail of the deterministic gather phase. Echoes exactly what the model was shown so Run History is self-explanatory. |
subjects | array | No | Per-subject verdicts, one per resolved subject in input order. Length 1 for a single subject; one per ticker when 'subject' is comma-separated (e.g. 'BTC,ETH,SOL'). The top-level sentiment/confidence/reasoning/marketContext mirror subjects[0]. Index subjects[i] to branch per ticker. |
model | string | No | The OpenRouter model that produced the classification. |
usage | object | No | OpenRouter token usage for the classification call. |
Examples
json{ "type": "ai-market-sentiment", "payload": { "text": "Workflow completed", "subject": "example-subject" }, "children": []}
bashcurl -X POST "https://api.b3os.org/v1/actions/ai-market-sentiment/test" \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "inputs": { "text": "Workflow completed", "subject": "example-subject" }}'
Payload fields can use workflow expressions such as {{$trigger.body.amount}}, {{$nodes.fetch.result.price}}, and {{$props.asset}} when the value should come from a trigger, prior node, or reusable workflow prop.