Open sourceMultiplayer AI and Apps

Shared Apps and Agents.

Your team, your agents, one context layer, scoped to each person.

Use it fromanywhere you work

Get started View on GitHub

What teams build on LemmaOpen a tile for the idea

Runson
Lemma.

Slack agents

# incidents# team-ops# deploys# product

# incidents

Alex

Database latency is spiking again!

Lemma Agent APP

I’ll take it from here.

Ops Team

Handoff complete. Monitoring.

Slack agents

Meet your team where it already works.

Incident coordinator
Standup digest
Knowledge teammate

See it in templates

Telegram agent + mini app

Lemma Agent bot

Hey! What can I help you with?

Status of deploy 2024-05-12?

Deploy 2024-05-12 Successfuldeploy-2024-05-12.log1.2 MB

Lemma Logbook

Open mini app

Connected
experience

Telegram agent + mini app

Chat is the surface. The pod is the system.

Deploy logbook
Shop mini app
Community concierge

See it in templates

Internal tools

NameStatusOwner

Onboarding flowIn progressMaya

API reliabilityAt riskJordan

Data pipelineHealthyLi

Access reviewIn progressSam

Q2 planningQueuedRiley

Internal tools

Build the exact tool you need, before lunch.

On-call board
Release checklist
Cost explorer

See it in templates

Skins for Codex

> codex --skin release-operator

Release pipeline

1. Build

2. Test

3. Deploy to stagingIn progress

4. Rollout to prod1 approval

Skins for Codex

Same agent. A new uniform for every job.

Release operator
Repo investigator
Pod builder

See it in templates

Account consoles

Acme CorpHealthy

Project health

Open requests

High

Medium

Account consoles

Open one client and see everything the team knows.

Project health
Request tracker
Milestone map

See it in templates

AI inboxes

Needs reply12

FYI8

Waiting5

Resolved24

AI inboxes

Wake up to an inbox that already triaged itself.

Triage labels
Drafted replies
Follow-up queue

See it in templates

Approval bots

New access request

ApproveReviseEscalate

Approval bots

Agents propose. Humans decide. Every call is logged.

Access requests
Change approvals
Escalation paths

See it in templates

What shared means

One of it, however many of you.

You share the agent itself, already running, so there is one of it to fix, one of it to improve, and one set of records underneath.

Send someone the link.
They open the same running agent.
Their work lands in the same records.
The agent itself, running once, for all of you.

And scoped to each person

PriyaApproves refunds, any amount
MarcoHis own jobs; refunds route to Priya
ClassifieragentReads tickets; read-only

Same system, different rights. One version of it, for everyone.

From a prompt to running team software

Build it, ship it, and hand it to your team.

Five layers. All of it real.

Build it where you already work.

Claude Code, Codex, Cursor, OpenCode, or inside Lemma itself. The agent authors the tables, agents, workflows, permissions, and the app, then verifies them through the same CLI.

Any coding agent
Authored as plain files
Verified by the agent that wrote it

The app and its agents go live together.

One URL for the people who use it. The agents start working on schedules, webhooks, and table events, and keep working while everyone is logged off.

Deployed at a URL
Agents run in the background
Versioned and rollback-able

Bring your team. And anyone else who needs it.

Teammates, clients, friends. They each get an invite link, the app, and their own account inside the pod.

Invite by link or email
Teammates, clients, guests
Humans and agents are both members

Decide exactly what each one can touch.

The same permission model covers people and agents. Grant a table, hide a resource, or require an approval before a consequential action goes through.

Per-table grants
Resource visibility
Approval gates on the risky steps

Use it from wherever you already are.

Telegram, Slack, WhatsApp, and email. Or open it as an app. Every entry point reads and writes the same records and respects the same permissions.

Chat surfaces
Your own UI or API
The app itself

Everything lives in a pod

Built from the ground up. One boundary around all of it.

Records first, then the functions over them, then the agents and workflows that move work, then the app people open. Permissions hold all of it.

Tables and files

What it remembers

tickets1,284 rows

id	uuid	primary key
customer	relation	→ customers
amount	money	indexed
status	enum	open · drafted · closed
created_at	timestamp	auto

refund-policy.mdmarkdown memory

## Thresholds

Refunds above $250 require an approver.

Repeat requests inside 30 days escalate to a lead.

Functions

What it does the same way every time

check_policydeterministic

fn check_policy(amount, customer) { policy = read("refund-policy.md") ifamount > policy.threshold return { requiresApproval: true }}

The predictable half

Same input, same output, every time. Plain code, end to end.
Validators, state transitions, and outbound actions live here.
Agents call them as tools, so judgment and rules stay separable.

Agents and workflows

What moves the work

Classifieragent

Can readtickets · customers · refund-policy.md

Can writetickets.status

Cannot touchrefunds · api_keys · connectors

Toolscheck_policy · search · table.update

refund-reviewworkflow

classifyagent

check_policyfunction

approvehuman

sendfunction

Apps and surfaces

Where people meet it

Support Opsdeployed

support-ops.lemma.work

tkt_418 Northwind · $420needs approval

tkt_417 Halden · $38sent

tkt_416 Ravel · $120sent

The same pod, other surfaces

Slack

ChatGPT

Claude

Gmail

API

Permissions and connectors

The boundary it all sits on

Accesspeople and agents alike

ticketsrefundsconnectors

DanaOwnereditapprovemanage

MayaMembereditrequestnone

ClassifierAgenteditnonenone

Connectorswhat the pod can reach

Stripebilling@northstar.coissue refunds

Gmailsupport@northstar.coread threads · create drafts

Surfaces

However the work arrives, it lands in the same pod.

Every surface knows who is asking, and writes to the same records.

Build it

The agent you already use builds the whole system.

The app, the tables, the agents, the workflows and the permissions. All written as files and checked by the same agent that wrote them.

✻ Claude Code~/support-ops

> Build a support ops pod: tickets, a refund policy, a classifier agent, and an approval before anything over $250 goes out.

✻ Designing the resource model…12s · 3.1k tokens

Writetables/tickets.yaml+34

Writeagents/classifier.yaml+21

Writeworkflows/refund-review.yaml+48 −2

Bashlemma pod import .✓

✓ Model tickets, customers, refunds
✓ Scope the classifier to tickets only
✓ Gate refunds over $250 behind an approver
◐ Verify the workflow pauses correctly

Ran refund-review with a $420 test ticket. Paused at approve, as expected.

Examples

Complete pods, running. Install one and make it yours.

Whole working systems, open source and running. Open one, see how it was built, make it yours.

Team operationsRoundtableA shared task board where people and agents can pick up work, leave a clear trail, and hand it back.Install and remix

WritingPaniniA writing space where agents comment on the exact passage, suggest an edit, and wait for you to decide.Install and remix

Customer supportFrontdeskA shared support inbox that sorts new messages, prepares an answer, and leaves the final send with a person.Install and remix

Personal productivitySmart InboxA personal inbox that follows the rules you choose and prepares replies without sending behind your back.Install and remix

Personal productivitySidekickA personal assistant that remembers how you work, handles recurring tasks, and asks before it acts outside the pod.Install and remix

DesignLemma DesignA place to turn a brief into several live prototypes, try them, and keep revising the direction that works.Install and remix

LearningNachiketaA learning space that starts with your goal, researches the subject, and changes the path as you attempt the work.Install and remix

CaptureDropA Telegram capture box that cleans up what you send and makes it possible to find again.Install and remix

Browse all templates

A pod is files. Export it, share it, remix it.

$ lemma pod export ./support-ops# the whole system, as files$ lemma pod import ./support-ops# ship it back, or anywhere else

Your machine.The whole stack runs on your laptop.
Your cloud.Self-host it, or use Lemma Cloud.
Your models.Your Claude or Codex subscription, Lemma-managed models, or any compatible endpoint.
Your code.AGPLv3 core. Apache-2.0 SDKs, CLI, and skills.

The agent you already use builds the whole system.

The app, the tables, the agents, the workflows and the permissions. All written as files and checked by the same agent that wrote them.

✻ Claude Code~/support-ops

> Build a support ops pod: tickets, a refund policy, a classifier agent, and an approval before anything over $250 goes out.

✻ Designing the resource model…12s · 3.1k tokens

Writetables/tickets.yaml+34

Writeagents/classifier.yaml+21

Writeworkflows/refund-review.yaml+48 −2

Bashlemma pod import .✓

✓ Model tickets, customers, refunds
✓ Scope the classifier to tickets only
✓ Gate refunds over $250 behind an approver
◐ Verify the workflow pauses correctly

Ran refund-review with a $420 test ticket. Paused at approve, as expected.

Shared Apps and Agents.

Slack agents

Telegram agent + mini app

Internal tools

Skins for Codex

Account consoles

AI inboxes

Approval bots

One of it, however many of you.

Build it, ship it, and hand it to your team.

Build it where you already work.

The app and its agents go live together.

Bring your team. And anyone else who needs it.

Decide exactly what each one can touch.

Use it from wherever you already are.

Built from the ground up. One boundary around all of it.

However the work arrives, it lands in the same pod.

The agent you already use builds the whole system.

Complete pods, running. Install one and make it yours.

Build it this afternoon. Have everyone using it by evening.

Shared Apps and Agents.

Slack agents

Telegram agent + mini app

Internal tools

Skins for Codex

Account consoles

AI inboxes

Approval bots

One of it, however many of you.

Build it, ship it, and hand it to your team.

Build it where you already work.

The app and its agents go live together.

Bring your team. And anyone else who needs it.

Decide exactly what each one can touch.

Use it from wherever you already are.

Built from the ground up. One boundary around all of it.

However the work arrives, it lands in the same pod.

The agent you already use builds the whole system.

Complete pods, running. Install one and make it yours.

Build it this afternoon. Have everyone using it by evening.