Abel AI
Causal time-series forecasting
表现
RD 熱度指數
RD 增長指數
比較
跟隨更新
跟隨列表
人物
關於Abel AI
Abel AI是一個專注於時間序列資料的因果AI平台。其核心業務是協助企業超越單純的預測,透過識別系統行為背後的因果驅動因素。該平台提供API和Python SDK,用於預測、異常偵測、根因分析與假設情境模擬。 公司的核心敘事是「從預測到因果」。傳統時間序列模型回答「會發生什麼」,而Abel AI則強調回答「為什麼會發生」和「如果我們介入會怎樣」。它建立包含變數、約束和延遲的即時因果圖,讓使用者能理解變化背後的機制,而不僅僅是觀察相關性。 Abel AI所解決的市場痛點是傳統預測模型的局限性:它們無法解釋為何做出某項預測,或系統將如何回應外部介入。這對於金融、營運及其他動態環境中的決策至關重要,在這些環境中,理解因果關係比單純的準確性更具價值。 在過去六個月(2026年3月至9月),Abel AI發佈了causal-abel,這是一個與Claude Code和Codex等AI程式設計代理整合的技能,使這些代理能夠直接執行因果分析。2026年3月,該公司分享了研究,證明自迴歸Transformer可透過梯度歸因內在地學習因果結構,其準確率幾乎是專門設計的因果發現方法的兩倍。2025年底,他們在NeurIPS舉辦了一場活動,展示因果AI如何超越黑箱預測,回答實際的「假設」問題,並開放了其平台的早期存取權限。
Between October 12 and 15, 2026, Andrey, the open-source causal discovery library from Abel AI Lab, shipped four releases in quick succession: 0.2.0 added hidden common cause methods, 0.3.0 brought in Tabu, XGES and hill-climbing search, 0.4.0 covered the pairwise ANM and PNL methods, and 0.4.1 fixed the DirectLiNGAM ordering implementation. The roadmap lists time-series methods as the work in progress for 0.5.0. On the product side, Abel packages forecasts as Decision Tokens for coding agents, with nine fields including prior, horizon, confidence, drivers, drift and watch signals, reachable over REST and MCP from Codex, Claude Code, OpenCode and OpenClaw. The underlying causal graph holds about six million nodes across more than 200,000 variables and refreshes daily, public portfolios track 110 paper-traded strategies, and the free developer tier allows 1,000 calls a month.
Abel AI's Future strategy and roadmap: Abel intends to extend its world-dynamics model from finance and macro into energy, industry and biology, building methodological credibility through the open-source library before monetising private enterprise deployments. The team caps confidence below 1.0 and requires every forecast to carry checkable watch signals.
Stephen Wang — Founder & CEO; former principal researcher at Snap and Meta, CTO at Amber Group, and researcher at Microsoft, specializing in deep learning, NLP, and temporal modeling.
Biwei Huang — CTO; also a professor at the University of California, San Diego, with research focused on causal discovery and machine learning.