TimeCopilot

The GenAI Forecasting Agent · LLMs × Foundation Time Series Models

[TimeCopilot @TimeCopilot]

and joint modeling of multiple series, and Azul previews what's coming to TimeCopilot.

Thank you, Geoffrey, for sharing the story behind the model with us. t0-alpha is available in TimeCopilot, and we can't wait to see what our community builds with it.

[TimeCopilot @TimeCopilot]

Watch our Co-founder and Co-CEO @azulgarza_ chat with @geoffnegiar, Co-founder and CEO of @forecastingco , in this episode of The Time Series. https://t.co/WTAq4ja49I

They talk about t0-alpha, The Forecasting Company's new time series foundation model, "early twice"

[TimeCopilot @TimeCopilot]

Geoffrey explains why they released it with open weights: to give back to the forecasting community and to let teams try the model for themselves instead of trusting an API.

Geoffrey also shares what's next for The Forecasting Company, including better use of covariates

[azul (she/her) @azulgarza_]

our model hub keeps growing! and today we're very happy to welcome t0-gamma, the newest addition to @forecastingco's family of time series foundation models! 😎

happy forecasting! 💙

🧵
#forecasting #timeseries #foundationforecast #timecopilot

[azul (she/her) @azulgarza_]

so excited to share that Timestral, the new time series foundation model by @MistralAI, is now available in FoundationForecast and @TimeCopilot! ✨

just last week, this new model was released, achieving state-of-the-art performance across GIFT-Eval, TIME, and fev-bench.

🧵

#forecasting #timeseries #foundation #timecopilot

[azul (she/her) @azulgarza_]

after launching @TimeCopilot, and engaging in conversations with Enterprise teams, and developers all around the world, we’ve heard frequently that many teams were keen to give a shot to time series foundation models, but often didn’t know where to start.

[azul (she/her) @azulgarza_]

we are very happy to officially release FoundationForecast, the unified open-source API for time series foundation models! ✨

🧵

[John Zeratsky @jazer]

Next week is our demo day for Character Labs!

We have room for a few more investors... if you are actively writing checks at inception/pre-seed and want to join us, DM me!

It's Tuesday, October 20 in downtown SF.

And here's a sample of the companies presenting:

• Ex-Stanford engineering team building GPU benchmarking and reliability optimization
• Ex-Caltech research team building whole brain models for animals and human
• Ex-Amazon geospatial experts building the AI harness for rapid land-use decisions
• @TimeCopilot, the open-source agent for accurate and complex time-series forecasts

[TimeCopilot @TimeCopilot]

https://t.co/Y8FOhSV8LX

[TimeCopilot @TimeCopilot]

Watch our Co-founder and Co-CEO, @azulgarza_ , chat with @LeventeZolyomi, PhD researcher at @nx_ai_com and co-author of the TiRex-2 paper, in this episode of The Time Series.

They explore how TiRex-2 goes beyond transformers with the xLSTM architecture, bringing new capabilities

[TimeCopilot @TimeCopilot]

and how that work could improve future time series foundation models.

Thank you, Levente, for sharing the ideas behind the model with us.

We’re excited to make TiRex-2 available in TimeCopilot and see what our community builds with it.

[Renée Rosillo 🧠✨ @rerosillo]

Agentic forecasting and TSFMs are hitting escape velocity in 2026. When my co-founder @Azul and her team started building TimeGPT, I don't think they knew they were starting a revolution for the time series field.

TimeGPT-1 came out in October 2023 as the first commercially available time series foundation model. Since then we've counted 63 more TSFM releases from Google, Amazon, Salesforce, IBM, Datadog, NXAI, Tsinghua and many others. Then last year the agentic wave started. Agents that pick, combine and reason over those models. The first agentic entry on the GIFT-Eval leaderboard was ours, @TimeCopilot, in July 2025. Today there are 32.

Inspired by NVIDIA's breakout-year chart for tabular foundation models, we created this for time series foundation models and for agentic forecasting.

If you work in time series, you know how fast this field moves. If you don't, this might be the hottest AI sector you've never heard of. Forecasting runs supply chains, energy grids, finance and infrastructure. The models behind it are being reinvented rapidly, and agents are making them usable for everyone.

Escape velocity is here. We're excited to be building in it, and grateful to everyone else who is too. 🚀

[azul (she/her) @azulgarza_]

why did @rerosillo and i start an agentic forecasting startup? 🤔

because we saw foundation models growing like never before, each with specific strengths across different kinds of data. whether companies and practitioners were working in electricity, healthcare, supply chain, finance, or infrastructure, we believed a layer on top was needed to select and orchestrate the best approach for each use case.

time series forecasting is a problem almost every company faces.

how are they going to grow? how much computational capacity should they allocate? how much inventory will they need? how many tokens are they going to burn? how will demand evolve?

companies evolve through time, and their numerical data is timestamped evidence of their existence. its the history that sets the ground for what comes next.

and just a few months after @TimeCopilot was released, agentic forecasting started becoming a category of its own. 😎

but what exactly is agentic forecasting?

we answered this question in detail at the International Symposium on Forecasting 2026 by the @IIForecasters.

here is the recording of an amazing experience we had: the opportunity to speak in front of some of the world's leading forecasters about agentic forecasting.

all my presentations are deeply rooted in who i am, who i want to become, the values i believe in, and of course where i believe the field is headed.

and this one couldn't be the exception.

yes-- we can be strong and beautiful together. 😉

happy forecasting! 💙

#forecasting #agenticforecasting

[TimeCopilot @TimeCopilot]

https://t.co/DFKUy3NJt8

[TimeCopilot @TimeCopilot]

We're on YouTube. Everything agentic forecasting, in one place, including The Time Series, our conversations with the researchers and leaders behind today's top forecasting models and companies.

You'll also find our research talks, model releases and more.

Come join us!

[azul (she/her) @azulgarza_]

what a blast of a conversation we had with Nick Erickson, creator of AutoGluon, for the new episode of The Time Series!

@innixma has carved out a unique path in machine learning. he has been devoted to open source for a long time, creating amazing projects such as AutoGluon. but not just that, his research has spanned different areas of machine learning: from tabular models and his work now at @prior_labs , to time series with models like Chronos-2, as well as benchmarks across both domains such as TabArena, BeyondArena, and fev-bench.

in this conversation, we deep dive into his journey, why building in the open is so relevant for producing high-quality software, and some of the ideas that have shaped his work.

we also shared a common perspective on how live benchmarks can be one of the strongest ways to keep pushing the limits of models forward. in this regard, we found a lot of common philosophy between Impermanent, our @TimeCopilot's live benchmark for time series forecasting, and TabArena.

hope you enjoy the conversation as much as we did, and happy forecasting! 💙

[TimeCopilot @TimeCopilot]

@azulgarza_ @forecastingco 🔥

[azul (she/her) @azulgarza_]

t0-beta by The Forecasting Company is now available in TimeCopilot! 🤘

since launching their foundation model t0-alpha earlier this year, trained on public and synthetic datasets, the @forecastingco team has continued upgrading its capabilities and has now introduced t0-beta, their new open-weights model.

this new release marks an improvement over the previous version. t0-beta shows comparable performance to Chronos-2, Tirex-2, and Toto-2 across point and probabilistic forecasting accuracy.

we are happy to welcome t0-beta to @TimeCopilot! the unified framework we are building helps companies and practitioners compare this model automatically with other foundation forecasting models under the same workflow. and now, they can do it with the t0 family as well. 😎

congrats to @geoffnegiar, @jodles89, and the rest of the TFC team on the release! 🫰

more info in the comments👇

[Renée Rosillo 🧠✨ @rerosillo]

@azulgarza_ we are cooking. So fired up and excited about what we are building @TimeCopilot.

Excited to connect with @mstockton thanks @jazer and @CharacterCap for bringing all of us together!