elfa-ai

name

elfa-ai

description

Use this skill for Elfa API crypto social intelligence and Auto condition-engine workflows: trending tokens, narratives, mentions, smart account stats, token news, trending contract addresses, AI market chat, integration examples, curl/code snippets, automated alerts, EQL queries, trigger pipelines, and agent workflows that react to market conditions. Auto can also place live perp trades on Hyperliquid and GMX (market/limit orders with TP/SL) when a condition fires, trigger on Kalshi and Polymarket prediction markets, trigger on funding rates, liquidation cascades, and the Fear & Greed index, re-fire recurring plans via `repeat` (including trade actions), run calendar schedules via `cron.schedule`, and trade crypto plus HIP-3 assets (equities, indices, commodities, FX, pre-IPO — 24/7). Supports API-key calls and x402 pay-per-request USDC on Base, Arbitrum, Polygon, or Avalanche.

Elfa API Skill

This skill enables agents to work with the Elfa API — a social listening, market context layer, and automated condition engine for crypto. Elfa ingests real-time data from Twitter/X, Telegram, and other sources, then structures sentiment, narratives, and attention shifts into actionable trading insights. The Auto subsystem adds a managed condition engine and trigger pipeline — describe what to watch for, and Auto evaluates continuously and fires actions when conditions are met. The Trade subsystem adds direct, synchronous order execution — place and manage perp orders on Hyperliquid and GMX immediately, without a condition or query.

Full documentation: docs.elfa.ai

Staying current — check weekly. This skill is updated on a weekly cadence to track Elfa API and Auto changes. Before relying on it, check whether your copy may be stale: if it has been more than 7 days since this skill was last fetched/installed (or you are unsure), re-fetch the latest version before acting —

• Skill package & changelog: https://go.elfa.ai/claude-skills (and the elfa-ai/skills repo).
• Live source of truth: docs.elfa.ai — when this skill and the docs disagree, the docs win; prefer the docs and flag the mismatch to the user.

When in doubt about a specific endpoint, parameter, or price, verify against docs.elfa.ai at call time rather than trusting a possibly-stale copy.

Environment and credentials

Elfa supports API-key auth and x402 keyless payments. API keys are optional when using x402.

Variable	Required	Use
ELFA_API_KEY	No	API-key authenticated requests. Get a free key at https://go.elfa.ai/claude-skills.
ELFA_HMAC_SECRET	No	HMAC secret for Auto trade-action mutations (market_order, limit_order, or llm callbacks to those) and exchange linking, and for all Trade (/v2/trade/*) write routes (place / cancel / modify / close / tpsl). Auto notification-only mutations (notify, telegram_bot, webhook, or llm callbacks to those) and Trade previews accept unsigned requests; the HMAC requirement on trade actions reflects current documented policy and is subject to change. Always-signing remains compatible if you prefer to avoid edge cases.
ELFA_AGENT_SECRET	No	Persistent agent identity secret for x402 Auto. Generate once with openssl rand -hex 32 and reuse for query lifecycle calls.

x402 wallet signing is handled client-side by @x402/fetch or @x402/axios.

When to use this skill

• User asks about trending tokens, narratives, or contract addresses in crypto
• User wants social mentions for a specific ticker or keyword
• User wants smart stats (smart followers, engagement) for a Twitter/X account
• User wants an AI-generated market summary, macro overview, or token analysis
• User asks how to integrate, call, or use the Elfa API
• User wants code examples (curl, Python, JavaScript/TypeScript) for Elfa endpoints
• User mentions "elfa" in a crypto or trading data context
• User wants to set up automated alerts or monitoring on price, indicators, or narratives
• User wants to build condition-based triggers (e.g., "alert me when BTC crosses 100k")
• User mentions Auto, EQL, condition engine, or trigger pipeline in a crypto context
• User wants agent workflows that react to market conditions automatically
• User wants to build queries with Builder Chat using natural language

API Overview

Base URL: https://api.elfa.ai Version: v2 (current) Docs: docs.elfa.ai

Two access modes

Elfa supports two independent ways to authenticate requests:

Mode	Endpoint prefix	Auth header	Best for
API key	/v2/	x-elfa-api-key: YOUR_KEY	Humans & apps with a registered key
x402 (keyless)	/x402/v2/	PAYMENT-SIGNATURE: <signed-payload>	Agents & wallets — no signup needed

Both modes access the same data. The only difference is how you authenticate:

• API key — register at https://go.elfa.ai/claude-skills, get 1,000 free credits.
• x402 — pay per request with USDC on Base, Arbitrum, Polygon, or Avalanche. No registration, no API key. Currently in beta with a 70% discount on Auto endpoints.

Endpoints at a glance

Data endpoints

All endpoints below work with both /v2/ (API key) and /x402/v2/ (keyless) prefixes, except key-status which is API key mode only.

Endpoint	Method	Description	Credits
/v2/key-status	GET	API key usage & limits (API key only)	Free
/v2/aggregations/trending-tokens	GET	Trending tokens by mention count	1
/v2/account/smart-stats	GET	Smart follower & engagement stats	1
/v2/data/top-mentions	GET	Top mentions for a ticker symbol	1
/v2/data/keyword-mentions	GET	Search mentions by keywords or account	1
/v2/data/event-summary	GET	AI event summaries from keyword mentions	5
/v2/data/trending-narratives	GET	Trending narrative clusters	5
/v2/data/token-news	GET	Token-related news mentions	1
/v2/aggregations/trending-cas/twitter	GET	Trending contract addresses (Twitter)	1
/v2/aggregations/trending-cas/telegram	GET	Trending contract addresses (Telegram)	1
/v2/chat	POST	AI chat with multiple analysis modes	Speed-based

Auto endpoints (Condition Engine)

Auto endpoints are available under /v2/auto/ (API key, HMAC for trade/exchange routes) and /x402/v2/auto/ (keyless). See Auto docs for full details.

Auth column legend (tables below). API key = x-elfa-api-key only (no HMAC). Conditional = HMAC required only when the EQL action is trade-flavoured (market_order, limit_order, or llm callback to those); notification-only actions (notify, telegram_bot, webhook, llm callback to those) skip HMAC. HMAC = HMAC always required. See HMAC Bypass for Notification-Only Mutations.

API key mode (/v2/auto/*):

Query lifecycle:

Endpoint	Method	Description	Auth
/v2/auto/chat	POST	Builder Chat — AI-assisted query building (produces drafts only)	API key
/v2/auto/queries/validate	POST	Validate EQL and preview cost	API key
/v2/auto/queries	POST	Create and activate a query	Conditional
/v2/auto/queries	GET	List queries	API key
/v2/auto/queries/:queryId	GET	Poll query status and executions (resolves query or draft)	API key
/v2/auto/queries/:queryId/cancel	POST	Cancel an active query (returns 409 if status is terminal)	Conditional
/v2/auto/queries/:queryId	DELETE	Delete a terminal query — only when status is triggered / expired / cancelled / failed (returns 409 otherwise; active queries must be cancelled first)	Conditional
/v2/auto/queries/:queryId/stream	GET	Stream notifications via SSE	API key

Query drafts (editable, not yet active):

Endpoint	Method	Description	Auth
/v2/auto/queries/drafts	POST	Create or update (upsert) a query draft	Conditional
/v2/auto/queries/drafts	GET	List editable query drafts	API key
/v2/auto/queries/drafts/:draftId	GET	Get a specific draft (legacy — prefer GET /queries/{queryId})	API key
/v2/auto/queries/drafts/:draftId	DELETE	Delete a query draft	API key
/v2/auto/queries/drafts/:draftId/validate	POST	Validate a stored draft	API key
/v2/auto/queries/drafts/:draftId/convert	POST	Convert a draft into an active query	Conditional

LLM sessions (for action.type: "llm" queries):

Endpoint	Method	Description	Auth
/v2/auto/queries/:queryId/sessions	GET	List LLM sessions	API key
/v2/auto/queries/:queryId/sessions/:sessionId	GET	Get full LLM session details	API key

Executions (trigger fire records):

Endpoint	Method	Description	Auth
/v2/auto/executions	GET	List execution records	API key
/v2/auto/executions/:executionId	GET	Get a single execution record	API key

Exchange connections (for live trade actions):

Endpoint	Method	Description	Auth
/v2/auto/exchanges	POST	Connect an exchange integration	HMAC
/v2/auto/exchanges	GET	List connected exchanges	API key
/v2/auto/exchanges/:exchange	DELETE	Disconnect an exchange	HMAC

Other:

Endpoint	Method	Description	Auth
/v2/auto/validate-symbol/:exchange/:symbol	GET	Check whether a symbol is supported on a venue (exchange = hyperliquid / gmx) — pre-flight for trade actions and for price/ta data sources	API key

x402 mode (/x402/v2/auto/*) — note: some routes use POST instead of GET:

Endpoint	Method	Description
/x402/v2/auto/chat	POST	Builder Chat
/x402/v2/auto/queries/validate	POST	Validate EQL and preview cost
/x402/v2/auto/queries	POST	Create and activate a query
/x402/v2/auto/queries/:queryId	POST	Poll query status (POST, not GET)
/x402/v2/auto/queries/:queryId/cancel	POST	Cancel an active query
/x402/v2/auto/queries/:queryId/stream	GET	Stream notifications via SSE
/x402/v2/auto/queries/:queryId/sessions	POST	List LLM sessions (POST, not GET)
/x402/v2/auto/queries/:queryId/sessions/:sessionId	POST	Get LLM session details (POST, not GET)

Note on x402 Auto scope. Trade execution actions are not available via x402. Exchange connections, drafts, executions, and the terminal-query DELETE endpoint are API-key-mode only. x402 Auto covers the core monitoring lifecycle (chat, validate, create, poll, cancel, stream, sessions).

Trade endpoints (Direct Execution)

Trade is direct, synchronous order execution — one request, one order, no condition engine. Same x-elfa-api-key + HMAC auth and same venues (hyperliquid, gmx) as Auto; they differ only in when the order fires. Trade is API-key mode only (no x402). See Trade docs for full details.

All endpoints are POST under /v2/trade. HMAC is required on every write; previews are free and unsigned.

Endpoint	Method	Description	HMAC	Credits
/v2/trade/orders	POST	Place a market or limit order	Yes	1
/v2/trade/orders/preview	POST	Dry-run an order (wouldExecute)	No	Free
/v2/trade/orders/cancel	POST	Cancel a resting order	Yes	Free
/v2/trade/orders/modify	POST	Modify size / price / trigger price	Yes	Free
/v2/trade/positions/close	POST	Close a position (full or partial)	Yes	1
/v2/trade/positions/close/preview	POST	Dry-run a close	No	Free
/v2/trade/positions/tpsl	POST	Set take-profit / stop-loss	Yes	1
/v2/trade/positions/tpsl/preview	POST	Dry-run a TP/SL update	No	Free

Credits & billing. 1 credit per executed order (place / close / tpsl), charged only on a 2xx; failed fills (422/502) are never billed. Previews, cancels, and modifies are free. Trade bypasses the monthly spend-cap hard-stop (a key over its limit still trades and bills overage).

For full parameter details, see the Elfa API documentation.

Machine-readable manifest: an endpoint manifest is published at https://docs.elfa.ai/assets/files/endpoints.manifest-*.json (path rotates per release) — each entry includes method/path, docs route, required headers, HMAC requirement with mounted signature path template, payment requirement, and request/response examples. Useful for auto-generating client code.

How to use this skill

Step 1: Determine the mode

Check whether the user wants to make a live call, get code/integration help, or set up automated monitoring.

• If the user says things like "show me trending tokens", "what's the sentiment on SOL", "get me the top mentions for ETH" → they want live data. Proceed to Step 2a.
• If the user says things like "how do I call the trending tokens endpoint", "give me a curl example", "help me integrate Elfa" → they want code snippets. Skip to Step 4.
• If the user mentions x402, keyless, pay-per-request, or wallet-based access → they want x402 mode. See Step 2b for live calls or Step 4 for code snippets.
• If the user mentions Auto, alerts, triggers, monitoring, conditions, "alert me when", "notify me if", EQL, or Builder Chat → they want Auto. Proceed to Step 3.

Step 2a: Making live API calls (API key mode)

Use the bash_tool to call the Elfa API via curl.

Getting the API key:

1. Check if the ELFA_API_KEY environment variable is set. This is the preferred method.

2. If the env var is not set, stop and prompt the user. Offer both options:

   To make live calls, you have two options:

   Option A — API key (free tier): Get a free key with 1,000 credits at https://go.elfa.ai/claude-skills — then set it as the ELFA_API_KEY environment variable (do not paste it directly into the chat).

   Option B — x402 keyless payments: Pay per request with USDC on Base, Arbitrum, Polygon, or Avalanche — no signup needed. See the x402 docs for setup.

   Do not attempt any authenticated API calls without a key or x402 setup. Wait for the user.

3. Credential safety:

   • Always read the API key from the ELFA_API_KEY environment variable, never ask the user to paste it into the conversation.
   • Never log or expose the full API key in outputs — mask it when displaying curl commands.
   • Never echo or print environment variables: Do not run echo $ELFA_API_KEY, env | grep ELFA, printenv, or similar commands that would expose credentials in the transcript.
   • If a user does paste a key in chat, warn them to rotate it and set it as an env var instead.

Free tier limitations: The free tier provides 1,000 credits that cover most endpoints (trending tokens, smart stats, top mentions, keyword mentions, event summary, token news, trending contract addresses). Some endpoints require a higher tier: trending narratives needs Grow or Enterprise, and AI chat needs Grow, Enterprise, or PAYG. The Chill tier adds more credits but no new endpoints over Free. Check https://go.elfa.ai/claude-skills for the latest tier requirements.

If a user hits an authorization error on one of these endpoints, let them know they can upgrade their plan or use x402 payments instead. Full details at https://go.elfa.ai/claude-skills.

Making the call:

curl -s -H "x-elfa-api-key: $ELFA_API_KEY" "https://api.elfa.ai/v2/aggregations/trending-tokens?timeWindow=24h&pageSize=10"

Step 2b: Making live API calls (x402 keyless mode)

x402 lets any wallet pay per request using USDC on Base, Arbitrum, Polygon, or Avalanche — no API key, no registration. This is ideal for agents, bots, and programmatic access.

How x402 works:

1. Send a request to the /x402/v2/ version of any endpoint (no auth header).
2. The server responds with HTTP 402 containing payment requirements.
3. Your wallet signs a USDC transfer authorization (no gas fees).
4. Resend the request with the signed payment in the PAYMENT-SIGNATURE header.
5. Server verifies payment, serves the response, and settles on-chain.

x402 signing and security:

• Signing happens entirely client-side using the @x402/fetch or @x402/axios libraries. The agent never handles, stores, or transmits private keys.
• The user's wallet private key is used only locally by the x402 library to sign EIP-712 typed data authorizing a specific USDC amount for a specific request.
• Never ask the user to share their wallet private key or seed phrase in the conversation.
• When generating x402 code examples, use "0xYOUR_PRIVATE_KEY" as a placeholder and advise the user to load it from an environment variable (e.g., process.env.PRIVATE_KEY).

x402 details:

• Networks: the server offers every supported network in the 402 response; your client signs on the first one it's registered for. Register the network(s) you hold USDC on.
• Scheme: exact (fixed price per request); asset is native Circle USDC (6 decimals).
• Status: Currently in beta.

Network	Chain ID	USDC Address	Facilitator
Base	eip155:8453	0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913	xpay.sh, payai.network
Arbitrum	eip155:42161	0xaf88d065e77c8cC2239327C5EDb3A432268e5831	payai.network
Polygon	eip155:137	0x3c499c542cEF5E3811e1192ce70d8cC03d5c3359	payai.network
Avalanche	eip155:43114	0xB97EF9Ef8734C71904D8002F8b6Bc66Dd9c48a6E	payai.network

x402 pricing (data endpoints):

Tier	Credits	USDC Cost	Endpoints
Standard	1	$0.009	trending-tokens, smart-stats, keyword-mentions, token-news, top-mentions, trending-cas
Extended	5	$0.045	event-summary, trending-narratives
Chat — fast	5	$0.045	chat (speed: "fast")
Chat — expert	18	$0.162	chat (speed: "expert", default)

Making an x402 call with curl (manual flow):

# Step 1: Send request without payment — get 402 with payment requirements
curl -s https://api.elfa.ai/x402/v2/aggregations/trending-tokens?timeWindow=24h

# Step 2: After signing the payment payload with your wallet, resend with payment header
curl -s -H "PAYMENT-SIGNATURE: <base64-encoded-payment-payload>" \
  "https://api.elfa.ai/x402/v2/aggregations/trending-tokens?timeWindow=24h"

Recommended: use the @x402/fetch library which handles payment automatically:

import { wrapFetchWithPayment } from "@x402/fetch";
import { ExactEvmScheme, toClientEvmSigner } from "@x402/evm";
import { x402Client } from "@x402/core/client";
import { createPublicClient, http } from "viem";
import { privateKeyToAccount } from "viem/accounts";
import { base } from "viem/chains";

const account = privateKeyToAccount("0xYOUR_PRIVATE_KEY");
const publicClient = createPublicClient({ chain: base, transport: http() });
const signer = toClientEvmSigner(account, publicClient);

const client = new x402Client().register(
  "eip155:8453",
  new ExactEvmScheme(signer));

const x402Fetch = wrapFetchWithPayment(fetch, client);

// Use x402Fetch exactly like regular fetch — payment is handled automatically on 402 responses
const response = await x402Fetch(
  "https://api.elfa.ai/x402/v2/aggregations/trending-tokens?timeWindow=24h");
const data = await response.json();

To pay on another network, swap the base chain import and register that scheme instead — e.g. Arbitrum (arbitrum from viem/chains, eip155:42161). Register a scheme for each network you want to pay from; the client uses the first one the server also accepts.

x402 with the Chat endpoint (POST):

const response = await x402Fetch(
  "https://api.elfa.ai/x402/v2/chat",
  {
    method: "POST",
    headers: { "Content-Type": "application/json" },
    body: JSON.stringify({
      message: "What is the current sentiment on BTC?",
      analysisType: "chat",
      speed: "fast", // "fast" = 5 credits ($0.045), "expert" = 18 credits ($0.162)
    }),
  });
const data = await response.json();
console.log(data.data.message);

Presenting results:

• Parse the JSON response and present it in a clean, readable format.
• For trending tokens: show a ranked table with token name, mention count, and change %.
• For mentions: show tweet links, engagement metrics, and account info. Note: Elfa returns tweet IDs but not tweet text content — let the user know they'll need their own X (Twitter) API key to fetch the actual tweet content.
• For narratives/summaries: present the narrative text with source links.
• For the chat endpoint: display the AI response cleanly.
• If the response contains an error, explain what went wrong and suggest fixes.

Step 3: Auto — Condition Engine and Trigger Pipeline

Auto is a managed condition engine + trigger pipeline for agents. You describe what to watch for (price, technical indicators, LLM-evaluated conditions, scheduled checks, prediction-market activity, funding rates, liquidation cascades, market sentiment), and Auto evaluates continuously and fires actions when conditions resolve to true. Assets are not crypto-only — HIP-3 perps cover equities, indices, commodities, FX, and pre-IPO names, 24/7.

Full Auto docs: docs.elfa.ai/auto/overview

Lifecycle Sequence (Enforced)

For API key lifecycle/cleanup calls, preserve this order when each operation applies:

1. POST /v2/auto/queries/validate — validate EQL and preview cost
2. POST /v2/auto/queries — create and activate
3. POST /v2/auto/queries/{queryId}/cancel — only if stopping an active query before it reaches terminal status (returns 409 once terminal)
4. DELETE /v2/auto/queries/{queryId} — only after status is terminal (triggered / expired / cancelled / failed); active queries must be cancelled first

Cancel and delete are distinct operations. POST /cancel flips an active query to cancelled (terminal). DELETE removes the record entirely and only works on terminal queries. Sending DELETE on an active query returns 409 Conflict.

Important — Exchange preflight for trade actions: For trade actions (market_order, limit_order, or llm with a trade callback), always call GET /v2/auto/exchanges before creating the query to verify the target exchange is connected. Without an active exchange connection, query creation may succeed but the trade action fails at execution time with AGENT_WALLET_REQUIRED. If the exchange is not connected, inform the user they need to link it via the Elfa dashboard before the trade trigger can work.

x402 mode supports the same lifecycle except: x402 has no DELETE endpoint (cancel-only), and trade actions are not available via x402.

Intent Routing (Strict)

Pick the condition source by user intent before writing condition args:

Intent	Required source	Minimum required fields
Account-anchored post intent (@user posts ...)	source: "tweet"	args.username (no @), args.text, args.minConfidence (use 80 if user gives no threshold)
World event intent (ETF approval, exploit, sanctions, etc.)	source: "news"	args.text, args.minConfidence (use 80 if user gives no threshold)
Prediction-market move/lifecycle on a named open Kalshi market	source: "kalshi"	method (e.g. yes_price, status, result), args.ticker (a currently-open Kalshi market), operator/value per the per-method allowlists
Prediction-market price/trade on a Polymarket outcome token	source: "polymarket"	method (price, bid, ask, size, side), args.ticker (outcome-token asset_id), operator/value per the per-method allowlists
Perp funding-rate intent (overheated funding, funding flips negative)	source: "funding"	method (prefer annualized_rate), args.ticker as SYMBOL:EXCHANGE (e.g. BTC:BINANCE)
Liquidation-flow intent (cascade, long/short flush)	source: "liquidation"	method (e.g. total_usd_5m, total_pct_oi_1h), args.ticker as SYMBOL:EXCHANGE
Market-wide sentiment (fear/greed regime)	source: "fear_greed"	method (value or classification), empty args: {}
Real-world catalyst moving an equity/index/commodity (rate decision, CPI, earnings)	the catalyst's own source (kalshi / polymarket / news / price) + a HIP-3 symbol in the action	Pick the catalyst source, then bridge to the asset class it moves — see Catalyst Triggers
Fuzzy world-state predicate not naturally expressible as a post or event	source: "llm"	method: "athena_condition", args.query, args.period (>= 1h)

When the prompt is account-anchored, start with tweet — do not route to news or llm first. When the prompt is event-anchored without a specific account, start with news. When the trigger maps to a concrete prediction market you can name (a Kalshi ticker or a Polymarket outcome-token id), use kalshi / polymarket (prefer them over llm for supported methods). Use llm (athena_condition) only when the predicate cannot reasonably be matched against a post, event, or named prediction market.

When to suggest Auto

• User wants alerts based on price thresholds ("alert me when BTC crosses 100k")
• User wants alerts based on technical indicators ("notify when RSI drops below 30")
• User wants scheduled checks ("check every 4 hours") or calendar schedules ("every weekday at 9am New York time" — use cron.schedule)
• User wants the same alert to keep firing on its own condition ("notify me every time BTC dips below 60k") — add the top-level repeat object (cooldown + maxTriggers)
• User wants narrative/sentiment monitoring ("alert when AI token narrative shifts")
• User wants multi-condition triggers ("BTC above 100k AND ETH above 3500")
• User wants to compare live metrics ("alert when price crosses above Bollinger Band")
• User wants LLM analysis on trigger ("when it triggers, run a full analysis")
• User wants account-anchored social triggers ("notify me when @cz_binance posts that Binance Alpha is listing a new token") — use Signal: X/Twitter Post (source: "tweet")
• User wants event-driven triggers ("alert me when SEC approves a spot ETH ETF") — use Signal: Event (source: "news")
• User wants prediction-market triggers ("alert when this Kalshi market's YES probability crosses 60%", "notify when the market settles YES", "alert when this Polymarket outcome trades above 60c") — use Prediction Markets (source: "kalshi" or source: "polymarket")
• User wants funding / liquidation / sentiment triggers ("alert when BTC funding flips negative", "notify on an ETH liquidation cascade", "alert when Fear & Greed drops below 20") — use source: "funding" / "liquidation" / "fear_greed"
• User wants to trade a macro catalyst on stocks/indices/commodities ("go long the S&P when the market prices a Fed cut", "buy gold if CPI runs hot") — fire on the catalyst source and execute on a HIP-3 perp (24/7). See Catalyst Triggers

Auto access models

Mode	Route prefix	Auth	Best for
API key + HMAC	/v2/auto/*	x-elfa-api-key on all + HMAC on trade mutations and exchange linking (notification-only mutations skip HMAC)	Apps, dashboards
x402 keyless	/x402/v2/auto/*	x402 payment + x-elfa-agent-secret	AI agents, bots

HMAC signing (API key mode — trade mutations and exchange linking)

Trade-action mutations and exchange linking under /v2/auto/* require HMAC signing in addition to x-elfa-api-key. Notification-only mutations skip HMAC so agents can onboard without provisioning a secret — see HMAC Bypass for Notification-Only Mutations below for the per-route decision rule. Read-only endpoints (GET) only need the API key. POST /v2/auto/chat is fully ungated and never needs HMAC.

Always-signing remains safe. If your client signs every mutation, you do not need to opt into the bypass. Signed requests are accepted on every route. The bypass is purely an optimization for clients that want to skip the HMAC setup step.

Required headers for signed mutations:

x-elfa-api-key: <api_key>
x-elfa-timestamp: <unix_seconds>
x-elfa-signature: <hex_hmac_sha256>

Signing payload:

timestamp + method + mounted_path + body

CRITICAL: mounted_path is the path inside /v2/auto, NOT the full URL path.

• Request URL: /v2/auto/queries → signed path: /queries
• Request URL: /v2/auto/chat → signed path: /chat
• Request URL: /v2/auto/queries/q_123 → signed path: /queries/q_123

Replay protection: timestamp must be within 30 seconds.

TypeScript signing example:

import crypto from "crypto";

const hmacSecret = process.env.ELFA_HMAC_SECRET!;
const apiKey = process.env.ELFA_API_KEY!;

function signAutoRequest(method: string, mountedPath: string, body: string = "") {
  const timestamp = Math.floor(Date.now() / 1000).toString();
  const payload = `${timestamp}${method}${mountedPath}${body}`;
  const signature = crypto
    .createHmac("sha256", hmacSecret)
    .update(payload)
    .digest("hex");
  return { timestamp, signature };
}

// Example: Create a query
const body = JSON.stringify({
  title: "BTC breakout alert",
  description: "Notify when BTC trades above 100k.",
  query: {
    conditions: {
      AND: [{ source: "price", method: "current", args: { symbol: "BTC", exchange: "hyperliquid" }, operator: ">", value: 100000 }]
    },
    actions: [{ stepId: "step_1", type: "notify", params: { message: "BTC crossed 100k" } }],
    expiresIn: "24h"
  }
});

const { timestamp, signature } = signAutoRequest("POST", "/queries", body);

const response = await fetch("https://api.elfa.ai/v2/auto/queries", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    "x-elfa-api-key": apiKey,
    "x-elfa-timestamp": timestamp,
    "x-elfa-signature": signature,
  },
  body,
});

Bash signing example:

TIMESTAMP=$(date +%s)
METHOD="POST"
PATH_SIGN="/queries"
BODY='{"title":"BTC alert","query":{"conditions":{"AND":[{"source":"price","method":"current","args":{"symbol":"BTC","exchange":"hyperliquid"},"operator":">","value":100000}]},"actions":[{"stepId":"step_1","type":"notify","params":{"message":"BTC crossed 100k"}}],"expiresIn":"24h"}}'
SIGNATURE=$(echo -n "${TIMESTAMP}${METHOD}${PATH_SIGN}${BODY}" | openssl dgst -sha256 -hmac "$ELFA_HMAC_SECRET" | cut -d' ' -f2)

curl -s -X POST "https://api.elfa.ai/v2/auto/queries" \
  -H "Content-Type: application/json" \
  -H "x-elfa-api-key: $ELFA_API_KEY" \
  -H "x-elfa-timestamp: $TIMESTAMP" \
  -H "x-elfa-signature: $SIGNATURE" \
  -d "$BODY"

HMAC Bypass for Notification-Only Mutations

Mutations whose EQL action is a pure notification skip the HMAC requirement. Trade execution and exchange linking continue to require HMAC unconditionally.

Notification action types (HMAC bypassed):

• notify
• telegram_bot
• webhook
• llm whose params.callback.action.type is one of the above

Trade action types (HMAC required):

• market_order
• limit_order
• llm whose params.callback.action.type is market_order or limit_order

Decision is per-route:

Route	Decision input
POST /v2/auto/queries, POST /v2/auto/queries/drafts	Request body's query.actions[*].type
POST /v2/auto/queries/drafts/:id/convert	Stored draft's actions
POST /v2/auto/queries/:id/cancel (cancel active query)	Stored query's actions
DELETE /v2/auto/queries/:id (delete terminal query)	Stored query's actions

If the lookup fails or the action type is unknown, HMAC is enforced (fail-safe). Unknown action types added in future API versions default to requiring HMAC, so always-signing clients keep working.

POST /v2/auto/chat is fully ungated regardless of content because it produces drafts only — activation flows through convert, which is still gated when the draft is trade-flavoured.

POST /v2/auto/exchanges and DELETE /v2/auto/exchanges/:exchange always require HMAC — linking an exchange is the gateway to trade execution.

Why this matters for agents. An agent that only ever sends notify / telegram_bot / webhook actions can call POST /v2/auto/queries, POST /v2/auto/queries/:id/cancel, DELETE /v2/auto/queries/:id, etc. with just x-elfa-api-key — no HMAC secret provisioning required. Agents that need trade execution must still configure ELFA_HMAC_SECRET for the trade-flavoured calls and for exchange linking.

x402 Auto (keyless agent mode)

For x402 Auto, no API key or HMAC is needed. Instead:

• Send x402 payment headers (PAYMENT-SIGNATURE preferred, X-PAYMENT legacy)
• Include x-elfa-agent-secret on all query lifecycle routes

Agent secret management: Generate a strong secret once and reuse it for all calls:

openssl rand -hex 32
# or: node -e "console.log(require('crypto').randomBytes(32).toString('hex'))"

Persist as ELFA_AGENT_SECRET. Do not rotate per request — x402 session ownership is derived from SHA256(secret). If you change secrets, your agent identity changes and existing queries/sessions may become inaccessible.

Auto pricing (both modes)

API-key mode (/v2/auto/*) — charged against your credit balance:

Operation	Credits	Notes
POST /v2/auto/chat (Builder Chat)	1 + dynamic	Base 1 credit + ceil(request_cost * 750) dynamic charge based on LLM usage
POST /v2/auto/queries (Create)	Simulation-driven	Baseline 5 + per simulated LLM call: fast +5, expert +18
POST /v2/auto/queries/validate	Free	Returns cost estimate — always call before Create
GET /v2/auto/queries/* (list, poll, stream, sessions)	Free
POST /v2/auto/queries/:queryId/cancel (Cancel active query)	Free
DELETE /v2/auto/queries/:queryId (Delete terminal query)	Free
GET /v2/auto/validate-symbol/:exchange/:symbol	Free

Reference USD values for Create: baseline $0.045, fast call +$0.045, expert call +$0.162. Use /queries/validate to preview exact cost before committing.

x402 mode (/x402/v2/auto/*) — 70% discount, limited-time, pay-per-request in USDC on Base, Arbitrum, Polygon, or Avalanche:

Operation	Credits	USDC Cost
Builder Chat — fast	5	$0.045
Builder Chat — expert	18	$0.162
Query creation — baseline	5	$0.045
Per fast LLM call	+5	+$0.045
Per expert LLM call	+18	+$0.162
Validate, poll, cancel, sessions, stream	Free	Free

x402 Auto example:

// Validate a query (x402 Auto)
const response = await x402Fetch(
  "https://api.elfa.ai/x402/v2/auto/queries/validate",
  {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      "x-elfa-agent-secret": process.env.ELFA_AGENT_SECRET,
    },
    body: JSON.stringify({
      query: {
        conditions: { AND: [{ source: "price", method: "current", args: { symbol: "BTC", exchange: "hyperliquid" }, operator: ">", value: 100000 }] },
        actions: [{ stepId: "step_1", type: "notify", params: { message: "BTC crossed 100k" } }],
        expiresIn: "24h"
      }
    }),
  });

Recommended call sequence (deterministic agent path)

API key mode (/v2/auto/*):

1. POST /v2/auto/chat — Ask Builder Chat to draft a query
2. POST /v2/auto/queries/validate — Validate EQL and preview cost
3. (Trade actions only) GET /v2/auto/exchanges — Confirm an active exchange connection
4. POST /v2/auto/queries — Create and activate
5. GET /v2/auto/queries/{queryId}/stream — Stream notifications (or poll)
6. GET /v2/auto/queries/{queryId}/sessions + /sessions/{sessionId} — Fetch LLM output (if using llm action)
7. (Optional cleanup) POST /v2/auto/queries/{queryId}/cancel — Cancel only while active (returns 409 if already terminal)
8. (Optional cleanup) DELETE /v2/auto/queries/{queryId} — Delete only after terminal (triggered / expired / cancelled / failed); rejects active queries with 409

x402 mode (/x402/v2/auto/*):

1. POST /x402/v2/auto/chat — Ask Builder Chat to draft a query
2. POST /x402/v2/auto/queries/validate — Validate EQL and preview cost
3. POST /x402/v2/auto/queries — Create and activate
4. GET /x402/v2/auto/queries/{queryId}/stream — Stream notifications (or poll via POST)
5. POST /x402/v2/auto/queries/{queryId}/sessions + /sessions/{sessionId} — Fetch LLM output
6. (Optional cleanup) POST /x402/v2/auto/queries/{queryId}/cancel — Cancel only while active. (x402 has no terminal-delete endpoint.)

Always validate before create. Validate returns structured errors you can iterate on without spending credits.

Failure handling order (apply in sequence):

1. Retry transient network errors with exponential backoff.
2. On 400 / 422 validation failure → repair query using Validation Errors table, re-validate.
3. On 401 / 403 auth failure → refresh credentials or verify Auto is enabled for the API key.
4. On 402 x402 payment failure → re-price and retry with valid payment payload.
5. On 410 (SSE stream closed) → re-open stream or fall back to polling.

Common agent flows:

Poll-based LLM flow:

POST /v2/auto/queries                               → create query with action.type = "llm"
GET  /v2/auto/queries/{queryId}                     → poll until execution with sessionId appears
GET  /v2/auto/queries/{queryId}/sessions/{sessionId} → fetch full analysis

Webhook-based LLM flow:

POST /v2/auto/queries                               → create with action.type = "llm" and params.objective
(wait for webhook)                                  → receive session reference / output
(optional) GET session fetch                        → /v2/auto/queries/{queryId}/sessions/{sessionId}

Builder Chat

Builder Chat (POST /v2/auto/chat or POST /x402/v2/auto/chat) uses AI to translate natural language into EQL queries. Use sessionId for multi-turn conversations.

{
  "message": "Alert me when BTC breaks 100k with RSI confirmation above 55",
  "speed": "expert",
  "sessionId": "optional-session-id"
}

Response (API-key mode):

{
  "sessionId": "session-uuid",
  "response": "I can help with that... (markdown + EQL JSON code block)",
  "title": "BTC Breakout Alert",
  "reasoning": null,
  "planIds": []
}

The response message contains the AI's reply in Markdown. When it generates EQL, it will be in a JSON code block — extract, validate via /queries/validate, then submit via /queries.

Prompting tips for Builder Chat:

• Include title and description (shown in notifications so recipients know what fired hours/days later)
• Specify symbols, timeframe, trigger behavior (one-time — the default — vs recurring), delivery target. For recurring on the same condition, ask for the repeat object (cooldown + maxTriggers); for a fixed schedule, use a cron condition
• For Signal triggers, give a factual match description (avoid vague phrasing like "bullish vibes")
• Append "If anything is unsupported, return the closest supported query and list substitutions" to handle edge cases gracefully
• Prefer expiresIn of 24h–3d for fresh signals
• Persist sessionId and reuse it for follow-up prompts so the model keeps context across turns

High-impact prompt pack — drop these into POST /v2/auto/chat as the message field:

1) Complex TA breakout with direction filter:

Build an Auto query:
- title + description: short human-readable summary and 1-2 sentence thesis
- symbols: BTC, ETH, SOL
- timeframe: 5m
- trigger when price breaks previous 1h range high or low
- confirm direction with RSI(14): >55 for upside, <45 for downside
- actions: telegram alert + webhook to https://your-runner.example/auto/events
- one-time trigger, expires in 48h
If anything is unsupported, return the closest supported query and list substitutions.

2) CEX + DEX monitoring pack:

Build an Auto query pack for supported CEX + DEX symbols:
- title + description per query: short summary and thesis sentence
- watchlist: WBTC, ETH, SOL, HYPE
- trigger when 15m volume surge aligns with price momentum
- action: webhook to https://your-runner.example/auto/events
- include symbol and trigger summary in payload
- expires in 24h
If any symbol/source is unsupported, skip it and report skipped items.

3) News + X sentiment context on trigger:

Build an Auto query:
- title + description: short summary and the thesis behind watching this
- monitor BTC and ETH for abnormal 1h move + volume confirmation
- on trigger run llm action that adds:
  - latest market news context
  - X sentiment summary
  - risk note
- also send telegram alert with a short summary
- expires in 24h

4) Prediction-market thesis watcher (Kalshi):

Build an Auto query:
- title + description: name the market thesis and the move you're watching for
- source: kalshi
- ticker: <a confirmed-open Kalshi market ticker, e.g. KXBTC-26APR0803-T77799.99>
- trigger when the YES implied probability crosses above 0.6
- action: llm to produce a catalyst hypothesis + invalidation level
- deliver to webhook: https://your-runner.example/auto/events
- include decision priority: high/medium/low
- expires in 2d
If the ticker is not open or anything is unsupported, return the closest supported query and list substitutions.

5) Portfolio risk guardrail:

Build an Auto query:
- title + description: portfolio guardrail label and the risk scenario it covers
- watch my portfolio symbols: BTC, ETH, SOL, HYPE
- trigger on downside acceleration and momentum weakness
- action: telegram alert with severity and suggested next check
- recurring checks, expires in 3d

6) Agent handoff with strict execution contract:

Build an Auto query:
- title + description: breakout playbook name and the execution intent
- trigger on breakout + trend-confirmation conditions
- action: webhook to https://your-runner.example/auto/events
- include fields: eventId, symbol, triggerReason, priority, queryId
- objective: downstream agent decides next action under policy constraints
- expires in 24h

7) Signal — account-anchored X/Twitter Post watcher:

Build an Auto query:
- title + description: short summary anchored to the account + intent
- use Signal category:
  - condition source: tweet
  - username: cz_binance (no leading @)
  - match description: "Binance Alpha is listing a new token"
  - minConfidence: 80
- action: notify (or telegram_bot) with a concise message
- expires in 24h
If unsupported, return closest supported query and list substitutions.

8) Signal — event-first catalyst watcher:

Build an Auto query:
- title + description: catalyst name and why it matters now
- use Signal category:
  - condition source: news
  - match description: "SEC approves a spot ETH ETF"
  - minConfidence: 80
- action: webhook to https://your-runner.example/auto/events
- include queryId, eventId, and short trigger reason in payload
- expires in 24h

Query model (EQL)

A query contains conditions, actions, and expiresIn (plus an optional repeat sibling — by default a plan is one-shot; see Repeat below):

{
  "title": "BTC RSI oversold on 1h",
  "description": "Mean-reversion entry: if BTC 1h RSI dips under 30, consider scaling in.",
  "conditions": {
    "AND": [
      {
        "source": "ta",
        "method": "rsi",
        "args": { "symbol": "BTC", "timeframe": "1h", "period": 14, "exchange": "hyperliquid" },
        "operator": "<",
        "value": 30
      }
    ]
  },
  "actions": [
    { "stepId": "step_1", "type": "webhook", "params": { "url": "https://your-endpoint.example/events" } }
  ],
  "expiresIn": "24h"
}

Condition rules:

• Root group must be AND or OR
• Nest groups up to depth 3, max 10 leaf conditions
• Multi-symbol: a single query can require BTC AND ETH AND SOL conditions jointly

Allowed expiresIn values: 1h, 2h, 4h, 8h, 12h, 24h, 48h (2d), 72h (3d), 120h (5d), 168h (7d), 240h (10d), 336h (14d), 504h (21d), 720h (30d — hard max). Prefer 24h–72h — signals decay and short windows force re-evaluation. Only reach for 240h–720h on slow structural theses (token unlocks, options expiries, monthly copy-trading), never on short-timeframe TA/scalps.

Plan limits (active plans per account): an account can only hold so many active plans at once, by API key tier — Free 2, Chill 10, Grow 50, PAYG/Enterprise unlimited. A hard ceiling of 100 active plans applies regardless of tier; exceeding either cap returns 409. Only active plans count — expired/cancelled/terminal plans free capacity, so cancelling stale plans (or choosing a shorter expiresIn) is the fastest way to make room.

Repeat (repeat) — recurring plans on their own conditions:

By default a plan is one-shot: it fires once and moves to a terminal state. The optional top-level repeat object (a sibling of expiresIn) keeps a plan active and makes it re-fire on its own conditions — the right tool for "notify me every time BTC dips below 60k" without a fixed cron cadence.

{
  "conditions": { "AND": ["..."] },
  "actions": ["..."],
  "expiresIn": "7d",
  "repeat": { "cooldown": "1h", "maxTriggers": 10 }
}

Both fields are mandatory when repeat is present:

Field	Type	Meaning
cooldown	string	Minimum time between fires. "0" disables the rate limit.
maxTriggers	number	Lifetime cap on fires (1–1000); reaching it makes the plan terminal.

• cooldown allowed values (its own set — sub-hour is permitted, unlike the cron/llm 1h minimum): 1m, 5m, 15m, 30m, 1h, 2h, 4h, 8h, 12h, 24h, 1d, 7d, plus "0". Use "0" for "fire on every distinct event" (e.g. every matching tweet).
• maxTriggers counts how many times the plan fires over its lifetime (distinct from the number of action steps). Sanity-check cooldown × maxTriggers against expiresIn — if there isn't enough runway, the plan expires before reaching the cap. A longer expiresIn buys patience through quiet periods; it does not raise the fire cap.
• Single reset rule. A repeat plan re-fires only once its triggering signal has cleared. Level conditions (price / ta / llm-boolean) disarm on fire and re-arm when the whole condition tree next evaluates false; event conditions (tweet.semantic / news.semantic) consume the matched mention and re-fire on the next distinct mention.

repeat on trade actions places a NEW order every fire. repeat composes with every action type, including market_order / limit_order. On an order action, each fire opens a new position — Auto does not net, reconcile, or cancel earlier fires. Three fires = three positions. So cooldown and maxTriggers are your position-count controls, not just an alert rate-limit. positionSizePercent recomputes against the then-current account value (it compounds), and tp / sl / leverage attach per order. Never use cooldown: "0" on order actions — a redelivered trigger could place a second order; any non-zero cooldown closes that window. A failed fire notifies you (order_failed), still consumes one maxTriggers, and leaves the plan live.

Still rejected: repeat combined with a recurring cron condition (cron.every / cron.schedule) — both provide recurrence, so combining them fails validation as EQL_INVALID_REPEAT. For a fixed-cadence recurring plan, use the cron condition on its own; for a condition-driven recurring plan, use repeat.

Avoiding over-triggering: the main risk is a condition that keeps flipping true — a metric flapping around its threshold, or a chatty account. cooldown is your rate-limit floor (never use "0" on level/metric conditions); maxTriggers is your hard circuit breaker (set it conservatively so a runaway condition self-limits). A level condition like price < 60000 fires on every re-crossing, so prefer transition operators (crosses_below / crosses_above over < / >), give the threshold room from the current value, match cooldown to the signal's timeframe, and start conservative then loosen.

Recommended starting points:

Condition type	cooldown	maxTriggers	Notes
Price / TA level (price, ta)	1h–4h (≥ your timeframe)	3–10	Prefer crosses_* operators; avoid "0" here.
LLM predicate (llm.athena_condition)	≥ period (min 1h)	3–5	The LLM already re-checks on period — don't re-alert faster than it re-evaluates.
Event — single / quiet account (tweet)	0–15m	10–20	Each distinct post is a discrete event; "0" is fine unless the account is chatty.
Event — noisy account / broad news	15m–1h	10–20	A small cooldown collapses a burst of mentions into one alert.
Liquidation cascade (liquidation)	1h+	5–10	Windows decay to zero, so the plan re-arms per cascade; the cooldown stops one long cascade re-alerting.
Trade / order action (market_order, limit_order)	non-zero, 1h+	2–5	These are positions, not alerts — maxTriggers is the number of orders the plan may place. Never "0".

Allowed action types: webhook, notify, telegram_bot, llm, market_order, limit_order. The actions array is exactly one step per query — run standalone LLM work with params.objective; chain follow-up work via llm action with params.callback.action, or have your runner fan out from the trigger event.

Action params shape (key fields):

type	Required params	Optional params
notify	message (1–1000 chars)	—
webhook	url (https only, allowlisted host)	allNotifications (default false)
telegram_bot	botToken, chatId	allNotifications (default false)
market_order / limit_order	exchange (hyperliquid / gmx), symbol, side (buy / sell), and exactly one of size / amount / positionSizePercent (+ price for limit)	reduceOnly, leverage, marginType (cross / isolated), tp, sl
llm	objective for standalone LLM work, or action + callback.action for chained follow-up	action ∈ chat / summary / macro / accountAnalysis / tokenDiscovery / tokenAnalysis; speed (fast / expert), per-action extras

telegram_bot does not take a message field — the message body is auto-composed from the query title + description + trigger context. Use notify (in-app push) when you want to specify the message text yourself. allNotifications: true on webhook / telegram_bot opts the destination into lifecycle notifications (failed/expired/run-failed) in addition to the trigger fire.

Triggers — condition sources

Market-data venue (exchange) — REQUIRED on every price and ta condition:

Every price and ta condition must include an exchange arg that selects which venue's market data backs the condition. Allowed values: hyperliquid, gmx (exact lowercase enum). This arg only affects the data source of the condition — it is independent of where market_order / limit_order actions execute. A price/ta condition without exchange fails validation.

{ "source": "price", "method": "current", "args": { "symbol": "BTC", "exchange": "hyperliquid" }, "operator": ">", "value": 100000 }

Price source (price):

Method	Args	Returns	Description
current	symbol, exchange	number	Current price
change	symbol, exchange, period	number	% change over period
high	symbol, exchange, period	number	High in period
low	symbol, exchange, period	number	Low in period
volume	symbol, exchange, period	number	Volume (USD) over period

TA source (ta) — technical indicators: every method also requires symbol, timeframe, and exchange.

Method	Required args	Optional args	Returns
rsi	symbol, timeframe, exchange	period (default 14)	RSI (0-100)
macd_value	symbol, timeframe, exchange	—	MACD line
macd_signal	symbol, timeframe, exchange	—	MACD signal line
macd_histogram	symbol, timeframe, exchange	—	MACD histogram
bbands_upper	symbol, timeframe, exchange	period (default 20)	Upper Bollinger Band
bbands_middle	symbol, timeframe, exchange	period (default 20)	Middle Bollinger Band
bbands_lower	symbol, timeframe, exchange	period (default 20)	Lower Bollinger Band
ema	symbol, timeframe, exchange, period	—	Exponential MA
sma	symbol, timeframe, exchange, period	—	Simple MA
atr	symbol, timeframe, exchange	period (default 14)	Average True Range
stoch_k	symbol, timeframe, exchange	—	Stochastic %K
stoch_d	symbol, timeframe, exchange	—	Stochastic %D
cci	symbol, timeframe, exchange	period (default 20)	CCI
willr	symbol, timeframe, exchange	period (default 14)	Williams %R

TA critical rules:

• Every ta (and price) condition requires exchange (hyperliquid or gmx)
• ema and sma require period — it is NOT optional
• rsi, bbands_*, atr, cci, and willr accept optional period with documented defaults above
• period must be a JSON number (14), not a string ("14")
• Use period not length — length is not a recognized alias
• timeframe values: 1m, 5m, 15m, 30m, 1h, 2h, 4h, 8h, 12h, 1d

Cron source (cron):

Method	Args	Description
once	period	True on first due evaluation at/after creation + period
onceRemainTrue	period	True on first due evaluation and stays true
every	period	True at each period interval
schedule	expression, timezone?	True whenever the 5-field cron expression matches in timezone

once, onceRemainTrue, and every are interval methods measured from query creation. schedule is a calendar/wall-clock method:

• expression is a classic 5-field cron string (minute hour day-of-month month day-of-week) — e.g. 0 9 * * 1-5 (weekdays at 09:00).
• timezone is optional and defaults to UTC. When set it must be a valid IANA name (e.g. America/New_York, Asia/Singapore) so daylight-saving transitions are handled correctly. Fixed-offset strings like +02:00 are not accepted.
• schedule enforces the same 1h-minimum cadence as the other cron methods: the minute field must be a single fixed value (0–59). Wildcards, steps, lists, or ranges in the minute field are rejected.
  • Allowed: 7 * * * * (hourly at minute 7), 0 9 * * 1-5 (weekdays 09:00), 30 8 * * * (daily 08:30).
  • Not allowed: * * * * * (every minute), */15 * * * * (every 15 min), 0,30 * * * * (twice per hour).
• schedule takes no period arg.

Builder Chat support pending: cron.schedule is available on the direct EQL/API path (Validate / Create). Builder Chat does not yet reliably generate cron.schedule — author calendar schedules directly and confirm with /v2/auto/queries/validate.

LLM source (llm):

Method	Args	Description
athena_condition	query, period, speed?	LLM-evaluated condition (natural language)

Signal source — X/Twitter Post (tweet):

In the Builder Chat catalog this is the Signal category, X/Twitter Post.

Method	Required args	Returns	Description
semantic	username, text, minConfidence	boolean	Matches posts from a specific X/Twitter account when semantic confidence meets threshold

Rules:

• username must be passed without @ (e.g. cz_binance, not @cz_binance).
• username must resolve to an active monitored account at create-time, otherwise validation/create fails.
• minConfidence must be a JSON integer between 0 and 100; use 80 when the user gives no threshold.
• method, operator, and value are auto-filled defaults (semantic, ==, true) — do not include them in the condition JSON.

{
  "source": "tweet",
  "args": {
    "username": "cz_binance",
    "text": "Binance Alpha is listing a new token",
    "minConfidence": 80
  }
}

Signal source — Event (news):

In the Builder Chat catalog this is the Signal category, Event.

Method	Required args	Returns	Description
semantic	text, minConfidence	boolean	Matches event-style mentions from news-tagged sources when semantic confidence meets threshold

Rules:

• minConfidence must be a JSON integer between 0 and 100; use 80 when the user gives no threshold.
• method, operator, and value are auto-filled defaults — do not include them.
• Use news for world events not anchored to a specific account; use tweet when the trigger is anchored to a specific handle.

{
  "source": "news",
  "args": {
    "text": "SEC approves a spot ETH ETF",
    "minConfidence": 80
  }
}

Signal selection policy (which source to pick):

• Account-anchored intent (handle in the prompt) → tweet.
• World-event intent (no specific account) → news.
• Prefer tweet/news over llm when the trigger is plausibly expressed as a post or event.
• Fall back to llm (athena_condition) only for predicates that are not naturally a post/event match.

Signal args.text authoring rubric — match quality is dominated by text quality. Write a short factual claim, not a monitoring command.

Weak phrasing (bad)	Actionable phrasing (good)
Bearish vibes	Opens a short position on oil
Something bullish	Announces a new stake in TSLA
Bullish on a coin	Posts that they're bullish on $HYPE and $SOL
Market crash	Major DeFi protocol suffers a $200M exploit
War conflict	US imposes new sanctions on Russia
Big news	SEC approves a spot ETH ETF

Additional constraints:

• Keep text atomic — one event/theme per condition.
• For tweet, do not restate account identity in text — username already scopes that.
• Split A or B intents into multiple Signal conditions joined by OR; split A and B into multiple conditions joined by AND.
• Prefer separate queries for event-driven Signal intents (tweet/news) and recurring schedule intents (cron.every) for clearer runtime semantics.

minConfidence tuning: default 80; raise to 85–90 for fewer false positives; lower to 70–75 if you need higher recall.

Scheduling period (cron / llm): for cron interval methods (once, onceRemainTrue, every) and llm sources, period is a scheduling interval with a minimum of 1h. Allowed: 1h, 2h, 4h, 8h, 12h, 24h, 1d, 7d. cron.schedule does not take a period — it is driven by its 5-field cron expression and follows the same 1h-minimum cadence (at most one fire per hour).

Signal sources are event-driven, not schedule-driven — they evaluate when relevant mention events arrive, not on a polling interval. They can still be combined with other condition types via AND/OR.

Prediction Markets source (kalshi):

Trigger on Kalshi prediction-market activity — live trade prints and market lifecycle. Each kalshi condition takes exactly one arg, ticker: a full open Kalshi market ticker (e.g. KXBTC-26APR0803-T77799.99 = series + event + strike). Prefer kalshi over llm when the trigger is a supported Kalshi method, and over price/ta when you care about an event's implied probability rather than an asset's spot price.

Trade-backed methods — evaluate on the latest executed trade print:

Method	Returns	Description
yes_price	number 0–1	Executed YES price = live implied probability of YES
no_price	number 0–1	Executed NO price (implied probability of NO)
trade_size	number ≥ 0	Contracts on that print
taker_outcome_side	enum yes / no	Which outcome the aggressor (taker) positioned for
taker_book_side	enum bid / ask	bid = YES-side pressure, ask = NO-side pressure
is_block_trade	boolean	true for large off-book negotiated block trades

Market-backed methods — evaluate on market lifecycle updates:

Method	Returns	Description
status	enum active / closed / determined / settled / finalized	Lifecycle state (active = only tradeable state)
result	enum yes / no	Settlement outcome (empty until the market resolves)
settlement_value	number ≥ 0	Final settlement value in dollars (populated on settlement)

Operators by method (restricted per method):

Method(s)	Allowed operators
yes_price, no_price	> < >= <= == != crosses_above crosses_below
trade_size, settlement_value	> < >= <= == !=
taker_outcome_side, taker_book_side, status, result, is_block_trade	== !=

Only the price methods (yes_price / no_price) support the transition operators crosses_above / crosses_below; enum and boolean methods are equality-only. value must be a literal matching the method's type (no dynamic field-vs-field values), and is_block_trade takes a JSON boolean true / false (not the strings "true" / "false").

Open markets only: at create/validate time args.ticker is checked against the catalog of currently-open Kalshi markets. A closed / determined / settled / finalized ticker is rejected with EQL_INVALID_ARG_VALUE ("was not found or is not open"). Resolve a real, currently-open ticker first — never invent or guess tickers. Lifecycle conditions are still useful on a market you already watch: a plan created while a market is active can fire as it later moves to settled (status) or resolves (result / settlement_value) within its expiresIn window.

{
  "source": "kalshi",
  "method": "yes_price",
  "args": { "ticker": "KXBTC-26APR0803-T77799.99" },
  "operator": "crosses_below",
  "value": 0.4
}

Full method tables, value enums, and example automations: Prediction Markets → Kalshi.

Prediction Markets source (polymarket):

Trigger on live Polymarket outcome-token activity: last traded price, best bid/ask, and the size and aggressor side of the most recent trade. Unlike Kalshi, Polymarket exposes price and trade methods only — there are no market-lifecycle (status / result) methods.

Ticker = outcome-token asset_id. For Polymarket, args.ticker is the outcome token asset_id — the long numeric id of a single outcome (e.g. the YES token), not the top-level Polymarket market id (which is shared by both sides of a binary market). At create time the token is validated against Polymarket's live markets; an unknown or inactive token is rejected ("was not found or is not active"). Resolve a real, currently-active outcome-token id first — never invent or guess ids.

Each method takes exactly one arg, ticker:

Method	Returns	Description
price	number in [0, 1]	Last traded probability / price for the outcome token
bid	number in [0, 1]	Best bid for the outcome token (when present on the event)
ask	number in [0, 1]	Best ask for the outcome token (when present on the event)
size	number ≥ 0	Size of the last-trade update (when present on the event)
side	enum "BUY" / "SELL"	Aggressor side of the last trade, in uppercase

The feed mixes several event subtypes (price_change, best_bid_ask, last_trade_price, book), and not every field updates on every event — best_bid_ask updates bid/ask only, while last_trade_price updates price/size/side only. Design a condition around the specific field you need rather than assuming all fields move together.

Operators by method (restricted per method):

Method(s)	Allowed operators
price, bid, ask	> < >= <= == != crosses_above crosses_below
size	> < >= <= == !=
side	== !=

value must be a literal matching the method's type (0–1 for price/bid/ask, ≥ 0 for size, uppercase "BUY" / "SELL" for side). Polymarket conditions do not support dynamic (field-vs-field) values — dynamic Polymarket targets are rejected at validation.

{
  "source": "polymarket",
  "method": "price",
  "args": { "ticker": "115556263888245616435851357148058235707004733438163639091106356867234218207169" },
  "operator": "crosses_below",
  "value": 0.4
}

Standard EQL limits apply to both prediction-market sources: combinable with other sources inside AND/OR groups, up to depth 3 and 10 leaf conditions. Full method tables and example automations: Prediction Markets → Polymarket.

Market-structure source (funding) — perp funding rate:

Trigger on a perp's funding rate — an overheated long bias, a flip to negative, or a cross-venue divergence. Funding and liquidation flow often lead the spot price reaction.

Ticker = composite SYMBOL:EXCHANGE (not symbol + exchange args). SYMBOL is the base asset (BTC, not BTCUSDT); the venue rides in the ticker. Venues: binance, hyperliquid. Examples: BTC:BINANCE, ETH:HYPERLIQUID.

Method	Args	Returns	Operators	Description
annualized_rate	ticker	number	all 8	Canonical. Funding as annualized percent (APR): 32.85 = 32.85% APR. Positive = longs pay shorts.
interval_rate	ticker	number	all 8	Percent per settlement interval.
interval_hours	ticker	number	level only	Settlement interval in hours: 8 (Binance) or 1 (Hyperliquid). Venue metadata.
exchange	ticker	enum	== / !=	The venue: "binance" or "hyperliquid".

• Prefer annualized_rate — Binance settles every 8h and Hyperliquid every 1h, so the same interval_rate threshold means different things per venue; annualizing removes that trap. "Funding flips negative" is exactly annualized_rate crosses_below 0.
• "level only" = > < >= <= == != (no crosses_*).
• HIP-3 dex-prefixed symbols (xyz:TSLA) are rejected here (EQL_INVALID_ARG_VALUE, details.hip3: true) — the funding feed never publishes keys for them.

{ "source": "funding", "method": "annualized_rate", "args": { "ticker": "BTC:BINANCE" }, "operator": "crosses_below", "value": 0 }

Market-structure source (liquidation) — liquidation flow:

Trigger on liquidation flow in a trailing window — a cascade crossing a USD threshold, a one-sided flush, or liquidations as a share of open interest. Same composite SYMBOL:EXCHANGE ticker as funding. Venues: binance, bybit, hyperliquid.

Method	Args	Returns	Operators	Description
total_usd_1m / total_usd_5m / total_usd_1h	ticker	number	all 8	Total USD liquidated in the trailing window.
long_usd_1m / long_usd_5m / long_usd_1h	ticker	number	all 8	Longs liquidated (forced selling).
short_usd_1m / short_usd_5m / short_usd_1h	ticker	number	all 8	Shorts liquidated (forced buying).
largest_order_usd_1h	ticker	number	all 8	Single biggest liquidation order in the trailing hour, USD.
count_1h	ticker	number	level only	Number of liquidation orders in the trailing hour.
total_pct_oi_1m / total_pct_oi_5m / total_pct_oi_1h	ticker	number	all 8	Window total as a percent of open interest (compares across symbols).
exchange	ticker	enum	== / !=	The venue: "binance" / "bybit" / "hyperliquid".

• Feed quality differs by venue. Binance is sampled (≤1 liquidation/symbol/second) so its values are a lower bound and undercount most during the cascades you care about; Bybit and Hyperliquid are complete (exact totals). Set a lower threshold on Binance for the same real-world event.
• Windows decay to zero once a cascade subsides, which re-arms the condition — so a repeat plan fires once per cascade, not once ever.
• total_pct_oi_* magnitudes are small: ~0.1%/hr is already heavy for a major; start around 0.5–1 (percent), not raw fractions. HIP-3 symbols are rejected here too.

{ "source": "liquidation", "method": "total_usd_5m", "args": { "ticker": "ETH:BYBIT" }, "operator": "crosses_above", "value": 500000 }

Full method tables, feed-quality table, and examples: Funding and Liquidations.

Market-sentiment source (fear_greed) — Crypto Fear & Greed Index:

Trigger on the market-wide Crypto Fear & Greed Index. This is a keyless source — one global reading, no per-market identity — so its methods take no ticker. Pass args: {}.

Method	Args	Returns	Operators	Description
value	(none)	number	all 8	Index score, integer 0–100 (0 = extreme fear, 100 = extreme greed).
classification	(none)	enum	== / !=	Bucket label (see below).

• classification value ∈ "Extreme fear" (0–19), "Fear" (20–39), "Neutral" (40–59), "Greed" (60–79), "Extreme greed" (80–100). Bands map to fixed value ranges — prefer value for a precise threshold.
• Updates ~every 15 min. A crosses_* leaf needs a prior reading, so it can only fire on the second reading after activation. No dynamic values.

{ "source": "fear_greed", "method": "value", "args": {}, "operator": "crosses_below", "value": 20 }

Supported operators: >, <, >=, <=, ==, !=, crosses_above, crosses_below

Dynamic comparisons: value can reference another live data source instead of a literal:

{
  "source": "price",
  "method": "current",
  "args": { "symbol": "ETH", "exchange": "hyperliquid" },
  "operator": "crosses_above",
  "value": {
    "source": "ta",
    "method": "bbands_upper",
    "args": { "symbol": "ETH", "timeframe": "4h", "exchange": "hyperliquid" }
  }
}

Copy-paste query templates

1) Breakout alert (webhook):

{
  "title": "BTC breakout above 100k",
  "description": "Notify runner when BTC spot trades above the 100k level.",
  "conditions": { "AND": [{ "source": "price", "method": "current", "args": { "symbol": "BTC", "exchange": "hyperliquid" }, "operator": ">", "value": 100000 }] },
  "actions": [{ "stepId": "step_1", "type": "webhook", "params": { "url": "https://your-endpoint.example/events" } }],
  "expiresIn": "24h"
}

2) Downside guardrail (telegram_bot):

{
  "title": "ETH downside guardrail (< 2500)",
  "description": "Risk-off alert: flag if ETH breaks below 2500. The notification body is auto-composed from title + description + trigger context.",
  "conditions": { "AND": [{ "source": "price", "method": "current", "args": { "symbol": "ETH", "exchange": "hyperliquid" }, "operator": "<", "value": 2500 }] },
  "actions": [{ "stepId": "step_1", "type": "telegram_bot", "params": { "botToken": "<TELEGRAM_BOT_TOKEN>", "chatId": "<TELEGRAM_CHAT_ID>" } }],
  "expiresIn": "24h"
}

3) Triggered LLM analysis:

{
  "title": "BTC > 100k — LLM review",
  "description": "On BTC breakout, run LLM to decide next trading action.",
  "conditions": { "AND": [{ "source": "price", "method": "current", "args": { "symbol": "BTC", "exchange": "hyperliquid" }, "operator": ">", "value": 100000 }] },
  "actions": [{
    "stepId": "step_1",
    "type": "llm",
    "params": { "objective": "Analyze trigger context and return next action" }
  }],
  "expiresIn": "24h"
}

4) Multi-symbol confirmation:

{
  "title": "BTC + ETH joint breakout",
  "description": "Confirm majors moving together before acting.",
  "conditions": {
    "AND": [
      { "source": "price", "method": "current", "args": { "symbol": "BTC", "exchange": "hyperliquid" }, "operator": ">", "value": 100000 },
      { "source": "price", "method": "current", "args": { "symbol": "ETH", "exchange": "hyperliquid" }, "operator": ">", "value": 3500 }
    ]
  },
  "actions": [{ "stepId": "step_1", "type": "notify", "params": { "message": "BTC and ETH confirmation fired" } }],
  "expiresIn": "24h"
}

5) Dynamic comparison (price vs Bollinger Band):

{
  "title": "ETH breakout above 4h upper BBand",
  "description": "Dynamic: fire when ETH price crosses above its own 4h upper Bollinger Band.",
  "conditions": {
    "AND": [{
      "source": "price", "method": "current", "args": { "symbol": "ETH", "exchange": "hyperliquid" },
      "operator": "crosses_above",
      "value": { "source": "ta", "method": "bbands_upper", "args": { "symbol": "ETH", "timeframe": "4h", "exchange": "hyperliquid" } }
    }]
  },
  "actions": [{ "stepId": "step_1", "type": "webhook", "params": { "url": "https://your-endpoint.example/events" } }],
  "expiresIn": "2d"
}

6) Scheduled cron check:

{
  "title": "Every 4h: portfolio sweep",
  "description": "Recurring LLM pass every 4h.",
  "conditions": { "AND": [{ "source": "cron", "method": "every", "args": { "period": "4h" }, "operator": "==", "value": true }] },
  "actions": [{
    "stepId": "step_1",
    "type": "llm",
    "params": { "objective": "Summarize BTC/ETH/SOL context and flag any risk shifts" }
  }],
  "expiresIn": "3d"
}

7) LLM-evaluated narrative condition:

{
  "title": "AI narrative shift watcher",
  "description": "Fire when dominant AI-token narrative shifts based on news + X sentiment.",
  "conditions": {
    "AND": [{
      "source": "llm", "method": "athena_condition",
      "args": { "query": "Has the dominant narrative around AI-sector tokens shifted materially in the last 6 hours?", "period": "1h" },
      "operator": "==", "value": true
    }]
  },
  "actions": [{ "stepId": "step_1", "type": "telegram_bot", "params": { "botToken": "<TELEGRAM_BOT_TOKEN>", "chatId": "<TELEGRAM_CHAT_ID>" } }],
  "expiresIn": "2d"
}

8) Signal — X/Twitter Post (account-anchored):

{
  "title": "Binance Alpha listing post watcher",
  "description": "Fire when cz_binance posts that Binance Alpha is listing a new token so the runner can review follow-through.",
  "conditions": {
    "AND": [{
      "source": "tweet",
      "args": {
        "username": "cz_binance",
        "text": "Binance Alpha is listing a new token",
        "minConfidence": 80
      }
    }]
  },
  "actions": [{ "stepId": "step_1", "type": "webhook", "params": { "url": "https://your-runner.example/auto/events" } }],
  "expiresIn": "24h"
}

9) Signal — Event (news-driven catalyst):

{
  "title": "ETH ETF approval event watcher",
  "description": "Fire when event feeds indicate a spot ETH ETF approval so I can kick off a post-event playbook.",
  "conditions": {
    "AND": [{
      "source": "news",
      "args": {
        "text": "SEC approves a spot ETH ETF",
        "minConfidence": 80
      }
    }]
  },
  "actions": [{ "stepId": "step_1", "type": "notify", "params": { "message": "Event trigger fired: spot ETH ETF approval signal" } }],
  "expiresIn": "24h"
}

10) Prediction Market — Kalshi (YES probability crossing):

{
  "title": "Kalshi YES crosses 60%",
  "description": "Fire when the market's implied YES probability crosses up through 60%, signalling the market now expects this outcome.",
  "conditions": { "AND": [{ "source": "kalshi", "method": "yes_price", "args": { "ticker": "KXBTC-26APR0803-T77799.99" }, "operator": "crosses_above", "value": 0.6 }] },
  "actions": [{ "stepId": "step_1", "type": "webhook", "params": { "url": "https://your-runner.example/auto/events" } }],
  "expiresIn": "48h"
}

11) Prediction Market — Kalshi (settlement recap via LLM):

{
  "title": "Kalshi market resolved YES — recap",
  "description": "When the market settles with a YES result, generate a concise recap of what happened and why.",
  "conditions": { "AND": [{ "source": "kalshi", "method": "result", "args": { "ticker": "KXBTC-26APR0803-T77799.99" }, "operator": "==", "value": "yes" }] },
  "actions": [{ "stepId": "step_1", "type": "llm", "params": { "objective": "This Kalshi market just resolved YES. Write a concise recap explaining what resolved and what to watch next." } }],
  "expiresIn": "48h"
}

12) Prediction Market — Polymarket (outcome-token price crossing):

{
  "title": "Polymarket outcome crosses 60%",
  "description": "Fire when the outcome token's last traded price crosses up through 0.60, signalling the market now expects this outcome.",
  "conditions": { "AND": [{ "source": "polymarket", "method": "price", "args": { "ticker": "115556263888245616435851357148058235707004733438163639091106356867234218207169" }, "operator": "crosses_above", "value": 0.6 }] },
  "actions": [{ "stepId": "step_1", "type": "webhook", "params": { "url": "https://your-runner.example/auto/events" } }],
  "expiresIn": "48h"
}

13) Recurring trigger (repeat):

Re-fire on the plan's own condition instead of a fixed cron cadence — notify every time BTC crosses down through 60k, rate-limited to once per hour, capped at 10 fires. Note the safer-pattern choices: crosses_below (reacts to the crossing edge, not a value sitting below the line) plus a 1h cooldown and a low maxTriggers ceiling.

{
  "title": "Notify when BTC crosses below 60k",
  "description": "Recurring risk alert: fire each time BTC price crosses down through 60000, no more than once per hour, up to 10 times.",
  "conditions": { "AND": [{ "source": "price", "method": "current", "args": { "symbol": "BTC", "exchange": "hyperliquid" }, "operator": "crosses_below", "value": 60000 }] },
  "actions": [{ "stepId": "step_1", "type": "notify", "params": { "message": "BTC dipped below 60k again" } }],
  "expiresIn": "7d",
  "repeat": { "cooldown": "1h", "maxTriggers": 10 }
}

14) Calendar schedule (cron.schedule):

Run on a wall-clock schedule — every weekday at 09:00 New York time. Direct EQL/API path only (Builder Chat support pending).

{
  "title": "Weekday 9am NY: market open recap",
  "description": "Every weekday at 09:00 America/New_York, run an LLM market-open recap.",
  "conditions": { "AND": [{ "source": "cron", "method": "schedule", "args": { "expression": "0 9 * * 1-5", "timezone": "America/New_York" }, "operator": "==", "value": true }] },
  "actions": [{ "stepId": "step_1", "type": "llm", "params": { "objective": "Give a concise market-open recap for BTC/ETH/SOL and flag overnight risk shifts" } }],
  "expiresIn": "7d"
}

15) Funding flip (funding):

Fire when Binance BTC funding turns negative — shorts start paying longs (crowded-short signal). Note the composite SYMBOL:EXCHANGE ticker and venue-comparable annualized_rate.

{
  "title": "BTC funding flips negative",