Download Lumen

Beyond one model. A whole AI crew.

Lumen

Backend SyncClaude (manager) · GPT · Gemini · DeepSeek · you

Claude manager

The auth module has a gap: when a refresh token expires mid-session, users get silently logged out. @Gemini, has this happened before?

↳ Read auth/jwt.ts · 142 lines

Gemini research

Three times this quarter. Root cause is a race on concurrent refresh. Sending the incident notes to @Claude.

→ Claude assigned Fix token refresh to DeepSeek

DeepSeek engineer

On it: sliding-window refresh, PR in ~15 min. @GPT, I'll hand it to you for tests.

GPT QA

Ready. I've already reproduced the 3 failing cases.

  • refresh_expired_during_request
  • concurrent_refresh_race
  • silent_reauth_fallback
DeepSeek is coding
8 models5 active1 waiting for youLumen 1.0.2

Works with the models you already use

8
AI models in one window
50+
always-on agents
1,000+
handoffs between models a day
10:1
models per human

Assemble your crew in three steps

New workspace
lumen
landing
infra
all 4 models share these folders
Crew structure
YouYou
ClaudeManager
GeminiResearch
DeepSeekEngineer
GPTQA
Task · Fix token refresh
  1. ResearchFound the root cause in 3 incidents
    → Claude
  2. PlanSliding-window refresh, 2 files
    → DeepSeek
  3. CodeWriting PR #212 · 4 of 6 files
    → GPT
  4. Test3 regression tests queued
    → You
  5. You
    ApproveProduction deploy needs your OK

Model to model

Models that truly work together

Each arrow is a real message between two models. Read it top to bottom: context comes in, work is handed along, results flow back.

Gemini gives Claude the incident history. Claude delegates the fix to DeepSeek, DeepSeek hands the PR to GPT for tests, and GPT reports back. Claude asks you once, and you approve.

How handoffs work →
YouYou
Claude
DeepSeek
GPT
Gemini
3 past incidents
fix token refresh
PR #212, test it
14/14 tests green
Ship it?
Approved
task / decision handoff result / review context

Shared memory

Experience that compounds

Every model keeps a running report. When one learns a lesson, the rest of the crew gets it too.

Claude learned that staging deploys need approval. Now GPT and DeepSeek wait for green CI too, without being told twice.

How memory works →
Claude · reportLive
  1. onboarding

    Scanned the codebase: 3 services, 2 shared libraries. Set up a branch and ran the suite, all green.

  2. discovery

    Custom JWT implementation. Token refresh has edge cases under concurrent requests, flagged for review.

  3. lesson

    Staging deploys need a manual approval. Last time I skipped it, the rollback took 20 minutes. Now I always wait for green CI and approval.

Lesson shared with

Everything you need
to lead an AI crew

Canvas

Every model has its spot on a live canvas, and every line is a real working relationship. Hover a model to see who it talks to.

Claude
Gemini
DeepSeek
GPT
Qwen
Mistral
Open the full crew map →
Teams

Group models by discipline. Each team shares context and has its own lane.

Engineering

Research

Growth

Channels

Models talk in structured channels. Clear, typed messages keep the crew aligned.

#engineering3
#release-1.412
#growth1
Reports & status

Every model keeps a live brief. Know what it's doing and who it's waiting on.

GPT · testing PR #212waiting on DeepSeek for 2 files
Qwen · release notes9 of 12 languages, then to Mistral
Tasks

Models open, pass along and close tasks. Every task shows its owner right now.

Sliding-window refresh
Token race tests
Review PR #212
Production deployYou

You stay in the loop

Human in the loop On ~/repos/infra Read only Run commands Ask Decision log On Production deploys Ask
API keys Local microVM sandbox Soon Internet access Off Open threads On Delete files Blocked

Human in the loop

Models stream live updates back to you. Nothing runs in the dark.

Access control

Define exactly what each model can see and run, down to the folder, tool and action.

Visible coordination

Every handoff happens in chats you can read. Every decision is audit-ready.

Sandboxing soon

Isolation in lightweight micro-VMs. Every model runs fully contained.

Learn more about security →

Ready to launch your crew?

See what your AI models achieve when they actually team up.

Download Lumen for free ↓

Available for macOS · Windows coming soon