Ygggoogle-research/timesfm
↗ GitHubTimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.
13,583
Stars
1,151
Forks
95
Watchers
180
Open Issues
Safety Rating A
The repository is a legitimate open-source research project from Google Research with an Apache 2.0 license. No hardcoded secrets, malicious code patterns, suspicious dependency configurations, or prompt injection attempts were detected. The codebase is consistent with an academic/research ML model release.
ℹAI-assisted review, not a professional security audit.
AI Analysis
TimesFM (Time Series Foundation Model) is a pretrained time-series forecasting foundation model developed by Google Research. It uses a decoder-only transformer architecture (200M parameters in v2.5) trained on large-scale real-world time series data, supporting context lengths up to 16k, continuous quantile forecasting, external regressors (XReg), and multiple backends (PyTorch and Flax). Released under Apache 2.0 and also available as an official Google BigQuery product.
Use Cases
- Zero-shot time-series forecasting without task-specific training
- Probabilistic/quantile forecasting for uncertainty estimation
- Financial, retail, and operational demand forecasting
- Integration into AI agent skill workflows for time-series analysis
- Fine-tuning or few-shot adaptation on domain-specific time-series data
Tags
Project Connections
RKiding/Awesome-finance-skills
The Awesome Finance Skills collection explicitly includes a time-series forecasting skill referencing the Kronos model; TimesFM could serve as an alternative or additional forecasting backend for those skill modules in financial agent workflows.
TauricResearch/TradingAgents
TradingAgents orchestrates multi-agent financial analysis; TimesFM could be integrated as the time-series forecasting component to produce price/demand predictions consumed by the trader and analyst agents.
rkiding/awesome-finance-skills
The finance skills collection provides agent-ready wrappers for forecasting tools; TimesFM's pretrained model can be wrapped as a SKILL.md-compatible forecasting backend for LLM coding agents.
karpathy/autoresearch
autoresearch provides an autonomous ML research loop; TimesFM's architecture and training methodology could serve as a target or baseline for automated time-series research experiments.
HKUDS/LightRAG
LightRAG handles knowledge retrieval; combining it with TimesFM could enable RAG-augmented forecasting systems where contextual documents (news, reports) inform or condition time-series predictions.