lemma-work/lemma-platform

Lemma Platform – an open‑source “AI teammate” for teams

What it is – Lemma is a framework that lets a team create, host, and interact with an AI‑powered “teammate” (called a pod). The teammate lives inside a shared, permission‑aware data store (tables, markdown files, workflows, etc.) and can be reached from the tools people already use – Slack, Microsoft Teams, WhatsApp, Telegram, or plain email. The core of Lemma is open‑source (AGPL‑v3 for the server, Apache‑2.0 for SDKs) and runs on the LLM subscription you already have (Claude Code, Codex, Cursor, OpenCode, or any OpenAI/Anthropic‑compatible API).

---

Key concepts

Concept	What it gives you
Pod	A self‑contained workspace that bundles tables, files, agents, workflows, permissions, and UI apps. Each pod can serve a whole team or a single user.
Agents	LLM‑driven workers with a defined role, tool access, and scoped permissions. They can read/write tables, call functions, and act on external connectors.
Workflows	Directed graphs mixing agents, code functions, loops, and human‑approval steps. Triggered by schedules, webhooks, table changes, or chat commands.
Tables & Files	Typed, queryable business data (row‑level security) plus markdown “memory” that stores policies, playbooks, notes, etc.
Apps & Pages	Low‑code UI surfaces that render live table views, widgets, and allow people to edit directly.
Permissions	Fine‑grained roles for both humans and agents – per‑table grants, resource visibility, and approval gates.
Surfaces	Slack, Teams, Telegram, WhatsApp, and email act as entry points. The same pod state is read/written regardless of where the request originates.

---

How you get started

1. Choose a host – Lemma Cloud (hosted by the project) or a local laptop installation via Lemma Desktop.
2. Install the CLI – uv tool install lemma-terminal (or the provided install.sh script).
3. Authenticate – lemma auth login connects the CLI to your chosen server.
4. Create a pod – lemma pod create support-ops --with-starter scaffolds a ready‑to‑use pod containing a sample table and an initial agent.
5. Talk to the teammate – lemma chat "what can you do in this pod?" or send a message from Slack/Teams/WhatsApp.

The README also lists a quick‑start for Lemma Cloud (single‑command bootstrap) and a local‑desktop flow that registers the desktop’s runtime endpoints with the CLI.

---

Main features

• Multiplayer AI – Multiple users can ask questions; the teammate tailors answers based on each user’s role and checks with the appropriate person before taking actions.
• Cross‑platform access – Same AI teammate works in Slack, Teams, WhatsApp, Telegram, email, or a custom web app.
• Shared & personal memory – Notes can be public to the whole pod or private to an individual; corrections are persisted.
• Low‑code apps – Type /visualize in a page to generate a chart, or embed live tables that auto‑update.
• Self‑improving harness – Corrections become permanent instructions, repeated patterns become reusable workflows, and new agents can be added without code changes.
• Provider‑agnostic – Uses your existing Claude Code, Codex, Cursor, OpenCode subscription, or any OpenAI/Anthropic‑compatible API key; also supports self‑hosted models via Ollama/LM Studio.
• Open‑source SDKs – Python (lemma-python) and TypeScript (lemma-typescript) libraries with React hooks for building custom front‑ends.

---

Typical use‑cases

Use‑case	How Lemma helps
Customer support – route tickets, draft replies, enforce approvals before sending.	Agents read ticket tables, suggest replies, wait for a human’s sign‑off, then send via the email connector.
Product feedback loops – collect, categorize, and track fixes.	A pod can ingest Slack/WhatsApp feedback, auto‑group themes, and surface a live board (the “Kit” example).
Personal productivity – unified inbox, rule‑based triage, reminders.	A single user’s pod stores personal notes, runs workflows that label incoming emails, and prompts before acting.
Team operations – shared runbooks, approvals, and status dashboards.	Tables hold run‑book steps, agents execute them, and the UI pages display live status with row‑level security.
Design & content creation – draft copy, get AI suggestions, keep versioned markdown memory.	Agents propose text, humans edit in a page, and the final document is stored as a markdown file with full history.

---

Extending Lemma

• Add new agents – Write a prompt file describing the role, grant it tool access, and import via the CLI.
• Create custom functions – Implement deterministic code (Python/TS) and expose it as a function that agents can call.
• Build bespoke apps – Deploy a static HTML page or a Vite React app with lemma apps deploy <pod> <path>; the app can read/write pod tables via the SDKs.
• Templates – The project ships ten ready‑made open‑source pods (e.g., Frontdesk, Smart Inbox, Panini) that can be cloned and customized.

---

Licensing & community

• Server core – AGPL‑v3, meaning you can run it on‑premises but must share modifications.
• SDKs – Apache‑2.0, permissive for building extensions.
• Documentation – Hosted at https://lemma.work/docs, with guides for installation, pod design, and SDK usage.

---

Bottom line

Lemma Platform is a genuine, open‑source framework for turning LLMs into collaborative, permission‑aware teammates that live inside a shared data “pod”. It abstracts away the plumbing (model API, memory, permissions, multi‑channel routing) so teams can focus on defining the tables, workflows, and UI that match their real‑world processes.