Salesforce xLAM
Type: full-code · Vendor: Salesforce · Language: N/A · License: cc-by-nc-4.0 · Status: active · Status in practice: emerging · First released: 2024-07-18
Salesforce xLAM is a family of Large Action Models tuned to plan and execute tasks by emitting structured function calls against available APIs rather than producing free-form text.
Description. xLAM is a family of Large Action Models from Salesforce AI Research designed to act as the action engine of AI agents. The models translate user intentions into executable actions and autonomously plan and execute tasks. The function-calling (fc) variants are optimised to return structured, JSON-formatted function calls based on the input query and the available APIs, comparable to the function-calling mode of hosted chat models. They are released for research under a CC BY-NC 4.0 license.
Agent loop shape. xLAM is the model at the core of an agent rather than a full runtime. Given a user query and a set of available APIs, an fc-series model produces a structured function call selecting the API and arguments to invoke; the surrounding harness executes the call and can feed the result back for the next step. The model is tuned so its action output follows a JSON function-calling format rather than free text, so downstream code can parse and dispatch the call deterministically.
Primary use cases
- function calling for AI agents
- translating user intent into executable API actions
- structured JSON tool-call generation
- autonomous task planning and execution
Key concepts
- Large Action Model (LAM) → large-action-models (docs) — An LLM tuned to translate user intentions into executable actions in the world, acting as the decision-making brain of an agent rather than a pure text generator.
- fc (function-calling) series → tool-use (docs) — Variants of xLAM optimised specifically for function-calling, returning fast, accurate, structured JSON tool calls selected from the query and the available APIs.
- JSON function-call output → structured-output (docs) — The action output format: parseable JSON that mirrors the function-calling mode of hosted chat models, so a harness can dispatch the call deterministically.
Patterns this full-code implements —
- ·Large Action Models (LAMs)
xLAM is a family of Large Action Models optimised for function calling and autonomous task execution rather than text generation, intended as the action engine of AI agents.
- ★★Tool Use
The models are trained for function-calling/tool use: given a natural-language query and a set of available functions, they produce the typed function/tool call(s) to execute rather than free-form te…
- ★★Structured Output
The fc-series models are tuned to return structured, JSON-formatted function calls based on the query and available APIs, so their action output conforms to a parseable schema rather than free text.
- ★★Plan-and-Execute
Beyond single calls, the LAMs are described as autonomously planning and executing multi-step tasks to reach a goal, with the prompt format inviting one or more function calls to achieve the purpose.
Neighbourhood
Click any neighbour to follow the lineage. Scroll to zoom, drag to pan.
Alternatives & relatives
References
Provenance
- Last analyzed:
- Last updated:
- Verification status: partial