Just-in-Time Forecasting
Generate predictions on-demand during conversations. No pre-training or model deployment required.
A standardized MCP framework enabling LLMs to reason over temporal signals with precision. Powered by RunnerTau models—specialized for time-series modality.
Our Mission
Unifying the fragmented time-series ecosystem under one agentic interface. Transforming how AI systems understand and reason about temporal data—making complex analysis as natural as asking a question.
Core Capabilities
Generate predictions on-demand during conversations. No pre-training or model deployment required.
Identify outliers and unusual patterns in real-time. Understand what deviates from expected behavior.
Trace anomalies to their origins. Correlate events across signals to understand the "why".
Transform raw signals into natural language insights tailored to your domain and context.
Multi-Modal: Text + SignalPurpose-built time-series models. Unlike general LLMs, RunnerTau natively understands temporal dependencies, seasonality, and signal dynamics.
Specialized ModalityCombine signals with images, documents, and structured data for context-aware insights.
Signal + Vision + TextInteractive Demo
Industries
Energy Sector
Monitor load patterns, predict renewable generation, and detect equipment anomalies before they cause outages.
Healthcare
From ECG arrhythmia detection to PPG-based vital monitoring. Enable clinical AI agents to reason about physiological signals.
Finance
Detect market anomalies, forecast trends, and analyze trading patterns with temporal reasoning.
IoT & Manufacturing
Predictive maintenance, quality monitoring, and process optimization for sensor streams.
Developer API
# Install TEMPO core and MCP server
pip install tempo-ts tempo-mcp
# Initialize with RunnerTau models
from tempo import TempoAgent
agent = TempoAgent(
model="runner-tau-pro", # or "runner-tau-base"
backend="auto"
)
Get started with a single pip install. TEMPO works with RunnerTau models or other MCP-compatible LLMs.
# Conversational forecasting
response = agent.query("""
What will the server load be next week?
Show me the 95% confidence interval.
""")
forecast = agent.forecast(
signal=load_data,
horizon=168,
confidence=0.95
)
Forecasts are generated just-in-time during conversations. No pre-training required.
# Detect anomalies conversationally
anomalies = agent.detect_anomalies(
signal=ecg_data,
method="isolation_forest",
sensitivity="high"
)
for anomaly in anomalies:
print(f"{anomaly.timestamp}: {anomaly.description}")
Multiple detection methods: statistical thresholds, isolation forests, or deep learning.
# Multi-modal analysis: signals + images + text
insights = agent.analyze(
signals=[ecg_signal, ppg_signal],
documents=[patient_report],
images=[chest_xray],
mode="cross_modal"
)
print(insights.correlations)
Combine time-series with images, documents, and structured data for comprehensive analysis.
# Start MCP server for agentic frameworks
from tempo_mcp import start_server
# Exposes: tempo_forecast, tempo_detect_anomalies,
# tempo_summarize, tempo_query, tempo_multimodal_analyze
start_server(host="localhost", port=8080)
Full MCP compliance means TEMPO works with Claude, GPT, and any agentic framework.