Salesforce AI Research
Confidence 0.75 · 2 sources · last confirmed 2026-09-04
The corporate research lab of Salesforce, and in this wiki the source of the xLAM open-weight “Large Action Model” family and the APIGen-MT synthetic-data pipeline that trains it — see 2025-04-04-prabhakar-salesforce-apigen-mt-xlam-2.
Notable for the wiki’s purposes because of a strategic asymmetry: Salesforce’s product side sells Agentforce, the agentic platform Sokolenko cites as an emblem of the agentic hype cycle (“Salesforce will try to sell you Agentforce”), while its research side open-sources small models and their training data — including the ones a practitioner uses to build an agent without buying any platform at all. Both things are true of the same company.
Researchers named in the corpus: Juan Carlos Niebles (Research Director, also Stanford), Silvio Savarese (Chief Scientist), Caiming Xiong, Shelby Heinecke, Akshara Prabhakar, Zuxin Liu.
Appears in this wiki via
- 2025-04-04-prabhakar-salesforce-apigen-mt-xlam-2 — APIGen-MT: Agentic Pipeline for Multi-Turn Data Generation via Simulated Agent-Human Interplay; the xLAM-2-fc-r model family.
Referenced by
- 2025-06-02-belcak-nvidia-small-language-models-future-agentic-ai — cites xLAM-2-8B as evidence that small models reach state-of-the-art tool calling.
- 2026-08-25-sokolenko-pycon-de-demystifying-agentic-ai-small-language-models — xLAM-2-32B is the model in the demo stack.