Salesforce releases xLAM-2: large action models extended to multi-turn agentic interactions, fully open-sourced
huan__wang · x · 2026-09-18
- Salesforce AI Research released xLAM-2, extending Large Action Models from function calling to multi-turn agentic interactions requiring coherence across whole conversations.
- Everything is public: the APIGen-MT-5k dataset (5k multi-turn dialogues with tool calls and agent policies), the earlier 60K dataset, the APIGen-MT synthesis pipeline, the ActionStudio training framework, papers, and a blog.
- The author stresses the team has open-sourced models, data, and research since the original xLAM for the community to build on.
More from coding & agent
- Dev discovers Codex security hardening switched persistent agent sessions to per-message instances — RileyRalmuto · 2026-09-18
- Polyphonic beta puts Claude Code, Codex and more agents in one Mac app, with local memory sharing — RileyRalmuto · 2026-09-18
- Redditor pitches AI app to scan road defects and auto-file mass claims against municipalities — InvestigatorIll3928 · 2026-09-18
- Shopify engineer builds Reflex: a Qwen-based browser experiment in structured decisions via WebGPU — multiply_matrix · 2026-09-18
- Baseten launches server-side web search for open models, claiming 15% lower latency — AccBalanced · 2026-09-18
- MCP Community Connect panel: non-tech adoption, registry challenges, and MCP in science and agriculture — digitarald · 2026-09-18