lemma-work/lemma-platform

What it solves

Lemma provides a "multiplayer" harness for AI agents, moving beyond single-user, single-session coding agents. It allows teams to build and share agents that operate on a shared context layer (pods) with persistent state, permissioned access, and the ability to integrate with common communication tools like Slack, WhatsApp, and Microsoft Teams.

How it works

Everything is organized into pods, which are self-contained environments containing:

• Tables and Files: Typed business data and Markdown memory for preferences and playbooks.
• Agents and Workflows: LLM workers with specific roles and tool grants, and graphs that mix agents, functions, and human approval steps.
• Functions: Predictable code for validators and state transitions.
• Permissions: Role-based access control for both humans and agents.
• Apps: Custom UIs built on pod APIs where teams can manage work and handle decisions.

Systems are built using existing coding agents (like Claude Code or Cursor) which author the pod's configuration as files, then import them via a CLI. These pods can be deployed locally, on a Lemma Cloud, or on a self-hosted server.

Who it’s for

• Teams needing shared AI agents that maintain state across sessions and integrate with their existing communication channels.
• Developers who want to use their preferred coding agents to architect and deploy full AI-driven business processes.
• Individuals looking for a personal assistant that keeps state and integrates with messaging apps.

Highlights

• Multi-surface integration: Connects agents to Slack, Microsoft Teams, Telegram, WhatsApp, and email.
• Agent-led development: Designed to be built by coding agents using a specialized set of skills.
• Human-in-the-loop: Built-in approval gates that pause workflows for human decision-making.
• Portable pods: Entire systems (data, agents, workflows) can be exported and imported as files.