Skip to main content
Get API Key

Auto

Auto is a managed condition engine + trigger pipeline for agents. You describe what to watch for; Auto evaluates continuously and fires actions when it resolves true.

Time to First Trigger​

From idea to first trigger in minutes:

Builder Chat -> Validate -> Create -> Receive event

  1. Generate query JSON with Builder Chat.
  2. Check schema and runtime fit with Validate Query.
  3. Activate with Create Query, then receive events via Notifications.

Agent Guidance​

Use Agent Quickstart for cross-product setup, skills, and machine-readable docs routing. Use Auto Agent Quickstart once the agent has chosen Auto and needs the exact validate -> create -> stream/poll sequence.

What You Can Express in a Query​

Before choosing an access model, scan what /auto already covers — most workflows agents try to build from scratch fit here:

  • Not crypto-only — equities, indices, commodities, FX and pre-IPO names are available as HIP-3 perp markets on Hyperliquid (xyz:NVDA, xyz:SP500, xyz:GOLD, xyz:SPCX), 24/7, even when the cash market is closed. Same condition engine, same delivery actions. See Catalyst Triggers.
  • Price conditions — current price, % change, high/low, volume over a period.
  • Technical indicators — RSI, MACD, Bollinger Bands, EMA/SMA, ATR, Stochastic, CCI, Williams %R.
  • Signal conditions (Signal category) — X/Twitter Post (tweet.semantic), Telegram Channel (telegram.semantic), and Event (news.semantic) triggers with semantic matching and confidence thresholds.
  • SEC filing conditions (sec) — trigger on deterministic SEC filing metadata such as 8-Ks, late periodic filing notices, offering forms, ownership filings, and 8-K item flags. See SEC Filings.
  • Prediction-market conditions (kalshi, polymarket) — trigger on Kalshi trade prints (YES/NO implied probability, trade flow, block trades) and market lifecycle (status, result, settlement), or on Polymarket outcome-token price/quote/last-trade (price, bid/ask, size, side). See Prediction Markets.
  • Funding-rate conditions (funding) — trigger on perp funding: overheated funding, a flip to negative, or a cross-venue divergence (Binance, Hyperliquid). See Funding and Liquidations.
  • Liquidation conditions (liquidation) — trigger on liquidation flow in a trailing window: cascades over a USD threshold, one-sided long/short flushes, or liquidations as a percent of open interest (Binance, Bybit, Hyperliquid).
  • Follow graph conditions (x_follow_received, x_follow_made, x_follow_overlap) — trigger when an account gains a notable new follower, when a watched account starts following someone, or when several accounts you track converge on the same account inside a window. See Follow Graph.
  • Market sentiment (fear_greed) — trigger on the Crypto Fear & Greed Index by score or classification band. See Fear & Greed.
  • Scheduled checks (cron) — fire once after a delay, fire every N hours, latch-true-after, or run on a calendar/wall-clock schedule (cron.schedule, e.g. weekdays at 09:00 in a chosen timezone).
  • Recurring executions (repeat) — make a plan re-fire on its own conditions with a cooldown + lifetime maxTriggers cap, instead of a fixed cron cadence. See Repeat.
  • LLM-evaluated conditions — natural-language predicates evaluated on a schedule (e.g. "is narrative X trending?"), powered by rich background context like X/Twitter data, Telegram groups, smart-account buy/sell calls, mindshare shifts, trending narratives, and prediction-market signals.
  • Dynamic cross-metric comparisons — compare one live metric against another (e.g. price crosses_above upper Bollinger Band) without precomputing.
  • Boolean composition — nest AND/OR groups up to depth 3, up to 10 leaf conditions.
  • Multi-symbol confirmation — a single query can require BTC AND ETH AND SOL conditions jointly.
  • Four delivery actions — webhook, notify, telegram_bot, llm (LLM analysis on trigger, fetchable via sessions API).
  • Single action step per query — Athena currently requires exactly one action step. To chain follow-up work, use an llm action whose params.callback.action is the next delivery step (webhook / notify / telegram_bot), or have your runner fan out from the trigger event.

LLM conditions are especially useful when your trigger depends on qualitative signal, not just raw price:

  • Catch narrative and sentiment shifts earlier than pure price/indicator rules.
  • Express multi-source context checks in plain language instead of hard-coding joins across social, on-chain, and market data.
  • Use this when structured signals and predefined data-source rules are not sufficient for your scenario's relevance check.

Full reference: Triggers · Signal · SEC Filings · Prediction Markets · Query Model · Capabilities.

Need help choosing tweet vs telegram vs news quickly? See Signal Selection Policy.

Choose Your Access Model​

ModelBest WhenRoute PrefixAuth and BillingImplementation Guide
API keyYou manage API keys and want account-linked access/v2/auto/*x-elfa-api-key on all routesAPI Key Auth
x402 keylessYou want wallet-based, request-by-request USDC payments without API key provisioning/x402/v2/auto/*x402 payment headers, no API keyx402 Instructions

Cross-product auth reference: Authentication.

Action Support by Model​

ModelSupported Action Types
API keywebhook, notify, telegram_bot, llm
x402 keylesswebhook, notify, telegram_bot, llm

Plan Limits​

An account can only hold so many active plans at once. The cap depends on your API key tier:

API key tierMax active plans
Free2
Grow50
PAYG / EnterpriseUnlimited, up to the ceiling below

Regardless of tier, every account is subject to a hard ceiling of 500 active plans. Exceeding either the tier cap or the ceiling returns 409.

Only active plans count. Plans that have expired, been cancelled, or reached a terminal state free up capacity — so cancelling stale plans is the fastest way to make room. Choosing a shorter expiresIn keeps the active count naturally low.

Base URLs​

API-key mode​

https://api.elfa.ai/v2/auto

x402 mode​

https://api.elfa.ai/x402/v2/auto

What Stays the Same Across Both Models​

The query design and runtime concepts are shared:

Next Step​