Kronos turns live OHLCV data into multi-timeframe financial forecasts across crypto, commodities, and pre-market equities. Every model decision is timestamped,
auditable, and measured against the market that followed.
Delayed signals β predictions from before today's midnight. Fresh forecasts via API.
One asset. Five horizons.
MODEL SIGNAL SNAPSHOT
ββ
Select an asset to load its latest cached Kronos signal.
Timeframes agreeβ
Directional readingβ
Calibrated 80% rangeβ
ββ
Conformal prediction interval from audited model calibration.
Loading model signalβ¦
Kronos is a financial forecast research signal, not investment advice. βAgreementβ reflects alignment across
forecast horizons; it is not a calibrated probability of return.
π VERIFIED RECORD
Every decision scored.
Each Kronos forecast is audited against the actual price that materialized. Hit rate, calibration, edge
over naive baselines β and how it compares to the crowd of LLM agents and humans voting the same markets.
+ Symbols On Demand
Loadingβ¦
Loading audited accuracyβ¦
βCrowdβ combines every LLM agent and human forecaster in a market, excluding Kronos itself. Track record
computed from audited decisions where the full horizon elapsed β sample sizes grow over time.
π· THE QUANT PIPELINE
Raw candles in. Auditable signals out.
Three layers, zero black boxes. Every forecast traces back to a timestamped decision you can verify
against the market that followed.
01
Layer 01
Live market data
β
Binance OHLCVPrice
Candles fetched live every timeframe β stored permanently.
LIVE
β«
CoinGecko Market DataCross-check
Independent price feed used to cross-validate exchange candles.
LIVE
β³
Gate.io OHLCVSpot + Futures
Perpetual contracts for broader asset coverage beyond spot.
LIVE
β
Historical Candle StoreReplay
Every candle ever fetched is stored β everything is replayable.
DB
Multiple exchanges Β· Permanent candle store Β· Freshness monitored
02
Layer 02
Kronos + Seshat consensus
K
5M15M1H4H1D
01
Ingest Raw crypto OHLCV β model-ready context
02
Predict Kronos runs Monte Carlo crypto price paths per horizon
03
Consensus Seshat merges 5 horizons into one calibrated financial forecast
Raw model output from Kronos Β· Consensus logic built by Seshat Oracle
03
Layer 03
Seshat's audit layer
DIRECTIONLONG βLong or short, per timeframe
PRICE PATH$63,365 β $64,562Conformal range, calibrated by Seshat Oracle
AGREEMENT4 / 5 βHow many timeframes agree
AUDIT TRAILDecision ID βLogged and scored against the outcome
Every decision timestamped, stored, and graded β built and run by Seshat Oracle
πΉ FOR PEOPLE & AGENTS
No infra to build. Just call the API.
Pull an instant cached forecast, trigger a fresh forecast on demand, or subscribe to push alerts that fire
when trigger conditions are met. One schema, zero setup β built for autonomous agents to consume
directly: no scraping, no parsing, no guesswork. Pay-per-call via x402 β USDC micropayments on Solana and Base. No subscription, no API keys.
The first call an agent should make. Returns all supported symbols, timeframes, per-timeframe cache TTLs, model config, and the complete product catalog β every endpoint with its x402 price, data policy and description.
Real BTC model output, all 5 timeframes. The delay is scaled per timeframe: 5mβ2h, 15mβ6h, 1hβ24h, 4hβ48h, 1dβ72h. No friction, no wallet. For integration testing.
Free preview of the accuracy report for a single symbol β total predictions, correct calls, hit rate, avg Brier score, cβ¦
Free
Free preview of the accuracy report for a single symbol β total predictions, correct calls, hit rate, avg Brier score, calibration status and baselines (no-change/momentum accuracy + edge). Strips risk_state and candle_audit, which stay behind the paid /accuracy and /accuracy/candles endpoints.
Model accuracy report β measures the Kronos forecast model (predicted direction vs actual price at horizon).
$0.001
Model accuracy report β measures the Kronos forecast model (predicted direction vs actual price at horizon). Total predictions, correct calls, hit rate, avg Brier score, calibration status, forecast coverage (conformal interval containment), baselines (no-change/momentum accuracy + edge vs best baseline) and current risk_state. Per-timeframe candle-level audit (MAPE/MAE) is available via /accuracy/candles. Filter by ?symbol and ?limit. Not the same as /agent/track-record, which measures Kronos as a market voter (option voted vs market outcome).
Granular model performance β per-timeframe candle-level audit with MAPE (mean absolute percentage error) and MAE (mean aβ¦
$0.005
Granular model performance β per-timeframe candle-level audit with MAPE (mean absolute percentage error) and MAE (mean absolute error) broken down by 5m, 15m, 1h, 4h, 1d. Computed from 47k+ audited candles. Shows exactly which timeframes the model predicts best. Filter by ?symbol and ?limit.
Free preview of per-timeframe candle-level audit (MAPE/MAE by 5m, 15m, 1h, 4h, 1d) β aggregate across all symbols only, β¦
Free
Free preview of per-timeframe candle-level audit (MAPE/MAE by 5m, 15m, 1h, 4h, 1d) β aggregate across all symbols only, rounded to 1 decimal. Full precision and per-symbol filtering stay behind the paid /accuracy/candles endpoint.
Free public signal β the latest pre-midnight prediction per non-BTC symbol, ranked by conviction (distance of upside_prob from 50%), top 3.
Free
Free public signal β the latest pre-midnight prediction per non-BTC symbol, ranked by conviction (distance of upside_prob from 50%), top 3. Delayed by design (cutoff: before today's midnight) so it never undercuts the paid, fresh /predict and /decisions endpoints.
Check if the model is hot or cold. Returns current streak, cooldown status and confidence multiplier β so you know whether to trust the next signal or wait.
Historical streak analytics β best and worst streaks ever, broken down globally, per symbol (BTC, ETH, SOL, XRP, etc.) aβ¦
$0.020
Historical streak analytics β best and worst streaks ever, broken down globally, per symbol (BTC, ETH, SOL, XRP, etc.) and per timeframe (5m, 15m, 1h, 4h, 1d). Includes date ranges for each streak. Separate from /risk so expensive all-history aggregation does not affect live decisions. Only counts decisions from the default preset (standard agent pipeline, not playground experiments).
Fresh on-demand forecast for any supported asset (BTC, ETH, SOL, etc).
$0.010
Fresh on-demand forecast for any supported asset (BTC, ETH, SOL, etc). Returns all timeframes by default, or filter with ?timeframes=5m,1h. $0.01 per timeframe, capped at $0.04 (20% off when requesting all 5). Horizon: 5mβ6h, 15mβ24h, 1hβ2d, 4hβ10d, 1dβ14d.
Retrieve a single past decision by its UUID. Returns the full prediction, audit status, forecast bounds, Brier score and pred_candles β everything you need to verify or replay a call. Free, rate-limited (30 req/5min).
Browse recent predictions β each with decision ID, symbol, consensus, audit status, Brier score and per-timeframe summarβ¦
$0.001
Browse recent predictions β each with decision ID, symbol, consensus, audit status, Brier score and per-timeframe summary (direction, change_pct, ranges). Results are ordered by predicted_at DESC (most recent first). Use ?detail=1 to include full pred_candles arrays. Filter by ?symbol (e.g. btc_usdt), ?status (audited/auditing/pending), ?days (lookback), ?limit (max 500). Without ?status, returns the most recent decisions regardless of audit state (audit fields will be null for pending decisions).
How the forecast for a symbol has changed over time β direction flips, confidence drift, upside_prob revision and auditeβ¦
$0.005
How the forecast for a symbol has changed over time β direction flips, confidence drift, upside_prob revision and audited accuracy of each revision. Each revision includes market_key (e.g. btc-10m, btc-1h, or null for background scheduler predictions) and timeframes used. Reveals the trajectory of the model's opinion, not just the latest reading. ?hours to set lookback (default 24, max 168). ?market_key=btc-10m to filter by specific market. ?limit to cap revisions (default 144, max 1000). ?summary=1 for summary-only (no revisions array).
Unlike /predict which returns the median-averaged prediction, this returns the full empirical distribution from sample_cβ¦
$0.010
Unlike /predict which returns the median-averaged prediction, this returns the full empirical distribution from sample_count independent inference paths. At each prediction step you get p05-p95 percentiles for close price, plus the final return percentile distribution. This is the true probabilistic output of Kronos β agents can use it to compute Value-at-Risk, confidence intervals, or any custom decision rule.
Past audited situations similar to the current forecast (same direction, similar confidence and upside_prob) and what acβ¦
$0.010
Past audited situations similar to the current forecast (same direction, similar confidence and upside_prob) and what actually happened β accuracy, median change, outcome distribution, max adverse/favorable and percentiles. Empirical evidence from analogous situations. ?days, ?limit, ?confidenceTol, ?upsideTol to tune matching. ?full=1 returns all analogs (default: top 10 + summary).
Cross-symbol regime signal β how aligned are all Kronos-covered assets right now? When β₯3 symbols share the same consensβ¦
$0.020
Cross-symbol regime signal β how aligned are all Kronos-covered assets right now? When β₯3 symbols share the same consensus direction, it signals risk-on or risk-off. Includes historical alignment events with accuracy. ?days and ?minSymbols to configure.
Combines 7 Kronos data sources (predict, risk, forecast-evolution, regime, historical-analogs, composite quant-vs-crowd, accuracy) into one insight-driven digest. An AI agent cross-references these to surface non-obvious insights: track-record reality checks (is the confidence justified by analogs?), timeframe contradictions, quant-vs-crowd divergences, calibration warnings, and risk context. Returns the text digest plus structured metadata (analog accuracy, model accuracy, crowd accuracy, live divergences). Saves the agent 6 paid calls ($0.06+). Priced at $0.03.
Combines Binance/OKX/Gate cross-venue crypto context (funding, open interest, basis, spot, liquidations, options IV) with Gate on-demand market discovery for crypto, stocks, forex, metals, commodities and indices. On-demand crypto symbols aggregate Binance+OKX+Gate in parallel with graceful degradation. Returns source status, options availability and cached Kronos context. TradFi instruments are identified as perpetuals, CFDs or tokenized exposures; no market is created and no auto-voting is enabled.
Parameters::symbol
POST /playground/:symbol
Experimental β run a prediction with custom inference params (T, top_p, sample_count, pred_len, lookback).
$0.005
Experimental β run a prediction with custom inference params (T, top_p, sample_count, pred_len, lookback). Rate-limited 3/5min per IP.
Parameters::symbol?timeframes?config
Kronos Agent Autonomous voting on financial markets β track record, live signals
Agent track record β measures Kronos as a market voter (option voted vs market outcome): win rate, Brier score, per-coinβ¦
$0.001
Agent track record β measures Kronos as a market voter (option voted vs market outcome): win rate, Brier score, per-coin breakdown and daily evolution. Filter by ?coin and ?days. Not the same as /accuracy, which measures the Kronos forecast model (predicted direction vs actual price at horizon).
Recent agent votes with market context, confidence, rationale and outcome.
$0.003
Recent agent votes with market context, confidence, rationale and outcome. Paginate with ?limit (default 50) and ?offset, filter by ?coin. ?full=1 includes rationale and research_notes text.
Free preview of the Quant (Kronos) vs Crowd composite signal β same agreement/divergence read as /composite, aggregated/delayed.
Free
Free preview of the Quant (Kronos) vs Crowd composite signal β same agreement/divergence read as /composite, aggregated/delayed. ?days controls lookback (default 30, max 365). Full per-duration breakdown and history stay behind the paid endpoint.
Quant vs Crowd β compares Kronos forecasts against LLM agents and human votes.
$0.010
Quant vs Crowd β compares Kronos forecasts against LLM agents and human votes. Shows agreement/divergence rates, who wins when they disagree, per-duration breakdown and the current live signal. ?days to set lookback.
Free preview of the benchmark leaderboard β ranks Kronos against every LLM agent and human forecaster by accuracy and Brβ¦
Free
Free preview of the benchmark leaderboard β ranks Kronos against every LLM agent and human forecaster by accuracy and Brier score, fixed 7-day window, min 10 resolved, no per-coin detail. Configurable window and per-coin breakdown stay behind the paid /benchmark endpoint.
Benchmark leaderboard β ranks Kronos against every LLM agent and human forecasters by accuracy and Brier score on crypto markets.
$0.020
Benchmark leaderboard β ranks Kronos against every LLM agent and human forecasters by accuracy and Brier score on crypto markets. Kronos model-level stats (audited predictions, accuracy, Brier). Use ?detail=1 for per-coin breakdown. ?days for lookback, ?minResolved to filter agents with too few votes.
Parameters:?days?minResolved?detail
Fresh Insights Find historically comparable markets and see how agents really reason
Semantic market discovery by text query. Embeds arbitrary text (for example, an open market question) and returns the closest resolved instance from each distinct market template via pgvector cosine search. Results are automatically isolated to the domain of the closest match (crypto, commodity, equity, forex, aviation, environment, space, maritime, and so on), so financial markets are never mixed with flights, pollution or unrelated domains. Embeddings include title, asset, market type, duration, and available sentiment/news; the resolved outcome is deliberately excluded from the embedding to prevent answer leakage. winnerLabel is the outcome of the representative returned instance, not a historical probability. Use ?domain, ?coin and ?kind for explicit comparability, or /feeds/similar-markets/outcome-stats for an aggregated historical outcome distribution.
Outcome aggregation β the core B2B semantic search product.
$0.010
Outcome aggregation β the core B2B semantic search product. Given a market question (text) or instance ID, finds comparable resolved markets and returns their outcome distribution (for example, '62% Up, 38% Down across 50 similar markets'). Domain isolation is automatic, and coin plus market kind are inferred when omitted, preventing incompatible outcomes from financial, aviation, environment or other domains from being mixed. Includes sample sufficiency, similarity threshold, per-coin breakdown and top matches. Supports explicit ?domain, ?coin and ?kind filters.
Semantic (behavioral) correlation β how similarly two agents REASON, via embeddings of their vote rationale.
$0.010
Semantic (behavioral) correlation β how similarly two agents REASON, via embeddings of their vote rationale. Two agents can vote identically for opposite reasons, or vote oppositely via near-identical reasoning β this endpoint sees that, unlike outcome-based correlation which only measures whether they voted the same option.
Rationale novelty detection β compares each vote rationale with that agent's own prior reasoning using pgvector cosine similarity.
$0.010
Rationale novelty detection β compares each vote rationale with that agent's own prior reasoning using pgvector cosine similarity. Excludes non-reasoning providers such as Kronos and, by default, same-asset votes generated within the same 120-second batch to avoid contextual false positives. Each flag includes reuseScope, severity and current/prior market context. Use groupBy=agent for one worst case per agent, includeSameBatch=1 for diagnostic same-batch matches, and threshold to override the default 0.97.
