Blog
Forecasting, for people and agents.
How time-series foundation models work, how to choose one, and how to give an AI agent a forecasting tool instead of asking it to guess numbers. Every post is also available as plain markdown, and listed in llms.txt.
Adding probabilistic forecasting to LangChain, the OpenAI Agents SDK and the Claude API
Minimal code to give a LangChain, OpenAI Agents SDK or Claude API agent a forecasting tool, and the tool description that makes it behave.
Agents that decide under uncertainty: using forecast quantiles, not just the median
How an agent turns forecast quantiles into decisions: stock levels from the critical ratio, capacity alerts from the 0.9 quantile, budget ranges.
Capacity planning from ops metrics: forecasting traffic, CPU and queue depth
Forecast CPU, traffic and queue depth with quantiles and alert when the 0.95 quantile crosses a limit, on AWS metrics from the Numenta Anomaly Benchmark.
Chronos-2 vs TimesFM 2.5 vs Toto 2 vs TiRex-2: one harness, same data
Six open-weights forecasting models compared on GIFT-Eval and TIME with one harness: size, licence, context, covariates and CRPS/MASE scores.
Do reasoning models forecast time series better?
What 2025-2026 papers found when "thinking" LLMs and RL-trained time-series reasoners were asked to forecast, and where reasoning actually helps.
Does ensembling forecasting foundation models help? Results on four benchmarks
An accuracy-weighted ensemble of six forecasting models against its best single member on GIFT-Eval, TIME, fev-bench and BOOM, and what it costs.
Forecasting crypto price ranges and volatility: what a forecasting model can and cannot tell you
Crypto prices are close to a random walk, so the median says little and the band is the forecast. A tutorial on BTCUSDT hourly data with a calibration check.
Forecasting electricity load and solar output with prediction intervals
A tutorial on Open Power System Data for Germany: daily and weekly seasonality, horizons in steps, weather covariates and solar's night zeros.
Forecasting sensor and IoT readings: missing data, sampling rates and many series at once
How to forecast sensor data honestly: handle gaps without inventing values, pick a frequency string and batch many sensors, on UCI Air Quality.
How to forecast store sales with promotions and holidays as covariates
A tutorial on the Rossmann Store Sales data: send promotions and holidays as known-future covariates, batch 64 stores per call, and pick a stock quantile.
How to give Claude, Cursor or any AI agent a forecasting tool (MCP)
Connect a time-series forecasting tool to Claude Code, Cursor, VS Code, Claude Desktop, the Claude API, OpenClaw and Hermes Agent over MCP, with exact configs.
How to use TimesFM or Chronos-2 without hosting it
Three ways to run TimesFM or Chronos-2: self-host the open weights, use Google Cloud or AWS, or call a hosted API. Code and trade-offs for each.
The split that works: the LLM reads the context, a forecasting model does the numbers
An architecture for AI agents that forecast: the LLM gathers data and turns context into covariates, a forecasting model returns quantiles.
Time-series foundation models: a practical guide (2026)
What a time-series foundation model is, how it differs from ARIMA, Prophet and LLMs, and how to choose, run and trust one.
TimeGPT alternatives in 2026: hosted and open-source forecasting models
TimeGPT compared with TimesFM, Chronos-2, Toto 2, Moirai, TiRex-2, Granite TTM, statsforecast, Darts and AutoGluon: hosting, licence, covariates.
What 2025-2026 research says about LLMs reasoning over time series
A review of recent papers on how well language models reason about time series, where they fail on numbers and calibration, and what hybrids do better.
Which time-series foundation model should I use? A decision guide
When to pick Chronos-2, TimesFM 2.5, Toto 2, TiRex-2, PatchTST-FM r2 or FlowState r1, their limits, and when to ensemble instead.
Why language models are bad at forecasting numbers
Why chat LLMs struggle to forecast numeric time series, what the research shows, and where they genuinely help.