How is a LAM different from an LLM?

An LLM generates language. A LAM uses a language model as its reasoning core but adds a planner, a tool/action layer, memory, and a perceive-plan-act-observe loop so it can actually get things done in the real world, not just describe how.

Is OpenLAM a Large Action Model?

OpenLAM is an open-source platform that runs an autonomous AI workforce built on the Large Action Model paradigm. It ships the LAM core loop (plan → act → observe → learn), a community-trained local model (OpenLAM 72B), a plugin/tool layer, governance, and a society of 100+ agents. It is free to self-host.

Can I run a LAM locally?

Yes. OpenLAM is open source and can be self-hosted with Docker, and it can use a local model (OpenLAM 72B via an Ollama-compatible endpoint) or bring-your-own provider, so a full Large Action Model workforce can run on your own infrastructure.

What is a Large Action Model (LAM)?

A Large Action Model (LAM) is an AI model that doesn't just generate text — it takes actions. Given a goal in plain language, a LAM plans the steps, calls the right tools and APIs, observes what happened, and keeps going until the job is done.

Definition

A Large Action Model (LAM) is an AI system that turns a natural-language goal into a sequence of real, executed actions. It uses a language model for reasoning, but wraps it in a planner, a tool/action layer, memory, and a feedback loop so it can operate software, call APIs, and complete multi-step work autonomously. The output of an LLM is text; the output of a LAM is completed work.

How a LAM works: the core loop

Every LAM runs some version of a perceive → plan → act → observe → learn loop:

Perceive — read the goal, the available tools, memory, and current state.
Plan — decompose the goal into an ordered set of concrete actions.
Act — execute one action: call a tool, an API, the browser, or another agent.
Observe — capture the real result and feed it back into the plan.
Learn — record what worked so future runs are faster and more reliable.

This loop is what separates an action model from a chat model. It is the engine behind OpenLAM's autonomous agents.

LAM vs LLM

An LLM is a component; a LAM is a system built around one. The LLM provides reasoning and language; the LAM adds the ability to do: a planner that sequences actions, a tool layer that connects to the outside world, memory that persists across steps, and governance that keeps risky actions in check. See the deeper breakdown in LAM vs LLM vs Agent.

How OpenLAM implements the LAM paradigm

OpenLAM is an open-source platform for running an autonomous AI workforce on the LAM paradigm:

The LAM core loop — plan, act, observe and learn, run by every agent.
A local model — OpenLAM 72B, a community-trained model served via an Ollama-compatible API, or bring your own provider.
A tool/plugin layer — connect CRMs, email, search, social, payments, MCP servers and more, so actions reach real systems.
A society of agents — 100+ agents organized into departments that delegate and collaborate.
Governance — approvals and guardrails for high-risk actions, plus honest reporting when a capability is missing.
Continuous learning — runs improve the workforce over time, with optional privacy-first federated learning.

What LAMs are used for