lumen --intro

Make AI think like the
right expert.

Lumen composes 181 expert roles with 192 stackable skills, persistent workspace memory, and any major AI provider into one workbench. The answer you get is shaped by the right expertise for the task, not by whoever the model met first.

Start 14-day free trial See what Lumen does

No credit card. Bring your own API key from Anthropic, OpenAI, Google, or xAI.

[ ROLES ]
181
[ SKILLS ]
192
[ PROVIDERS ]
4
[ ON TRIAL ]
[ STATUS ]
online
lumen / workspace: payments-refactor opus 4.7
YOU Help me refactor the payment processor. Stripe webhook handler is doing too much; I want to split signature verification from event dispatch and add idempotency.
LUMEN (role: Coder, skill: API Refactor)
Copy
Refactor apps/web/functions/api/stripe/webhook.ts.

Split into three layers:
  1. verifyStripeSignature(req, secret) -> {ok, event} | {error}
     Pure function. Returns the parsed event on success.
  2. dispatchEvent(event, db) -> {handled, ignored, error}
     Switch on event.type, delegate to per-type handlers.
  3. recordOnce(event.id, db) -> {first, duplicate}
     Insert into a new processed_events(id PRIMARY KEY, processed_at)
     table; if INSERT conflicts on id, return duplicate=true and skip
     dispatch. Idempotency.

Constraints:
  - Keep the existing 400 / 503 / 500 response shapes.
  - Do not import Stripe SDK; use fetch + WebCrypto as the current file does.
  - Add a schema migration for processed_events.

Paste this into Claude Code, Cursor, or ChatGPT. Lumen drafts; your AI agent runs.

This is what every Lumen turn looks like. Try it free →

//What people use Lumen for

Six concrete role+skill combos, each with the intent someone would actually type.

CoderCode reviewRefactor

Refactor a legacy module

> Split the auth middleware into signature verification + dispatch + idempotency. Keep API stable.

Lumen returns a 3-step prompt mapping the refactor to specific files, with constraints carried through.

PentesterOWASP Top 10Threat model

Audit auth for OWASP issues

> Review apps/web/functions/api/login.ts for OWASP A1-A10 risk. Rank findings by severity.

Lumen returns a prompt that maps to OWASP categories and asks for severity-ranked, fix-included findings.

ArchitectSpec writerData model

Plan a real-time feature

> Add live notifications when a customer's prompt fence is ready. Sketch API + data model + rollout.

Lumen returns a prompt with sections for API design, data model, infra changes, and a phased rollout plan.

DebuggerStack readerMin-diff fix

Isolate a production bug

> Customers see ERR_MODULE_NOT_FOUND on cold restart. Walk through the stack, propose min-diff fix.

Lumen returns a prompt that asks for a step-by-step trace + the smallest possible change to fix it.

Tech writerAPI referencePlain language

Turn notes into a doc

> Convert these rough notes into a README for the cron worker. Audience: new engineer joining the team.

Lumen returns a prompt structured for a README: setup, deploy, schedule, troubleshooting, related docs.

NegotiatorBATNAPrincipled

Prep a vendor call

> Brief me for the call with Datadog tomorrow. Goal: cut MRR by 30%. Their BATNA, mine, openers.

Lumen returns a prompt for a BATNA / ZOPA briefing with concrete openers and a fallback ladder.

Pick any of the 181 roles, stack any of the 192 skills. Try it free →

//What Lumen actually does

Not a chatbot. A composition layer that turns any AI provider into a domain expert with memory.

Composable expertise

181 expert roles times 192 stackable skills equals millions of expertise combinations. Coder + Code Review + Type Safety Guard for a refactor. Pentester + OWASP Top Ten + Threat Modeling for a security pass. Negotiation Coach + BATNA + Principled Negotiation for a vendor call. Pick the lens, stack the discipline, the answer specializes.

Workspace memory

Lumen remembers what you have been working on across sessions. Directives you set, patterns it has learned, episodic context from prior turns. Hybrid keyword and vector retrieval surfaces what is relevant to your current intent. Most AI forgets every conversation; Lumen builds knowledge over time.

Multi-provider routing

Pick a provider per workspace, not per account. Claude Sonnet 4.6 for tool use and coding. o3 for pure reasoning. Gemini 2.5 Flash for cheap bulk work. Grok for real-time data. Your key, the provider's bill, Lumen's routing.

Blueprint Designer

14-section product spec tool that turns a conversation into a shipped doc bundle: README, architecture, API reference, data model, decisions. The blueprint stays a living artifact your AI sees on every turn; the rendered docs land where your team works.

Coding Mode

Flip a workspace to Coding Mode and Lumen auto-picks the best-fit role per turn from a curated pool of 25 software-focused roles. Coder, Tech Lead, DevOps Engineer, Security Auditor, ML Engineer. No fiddling with role pickers between turns; the right lens applies itself.

Two-pane workflow

Type intent on the left, get a ready-to-paste prompt. Run it in Claude Code, ChatGPT, Cursor, or whatever AI agent you trust. Paste the output back to refine. Lumen drafts, the agent executes, you stay in control of what runs and on what context. Tight loop, no chat sprawl.

Tenant isolation

Every workspace, every chat, every API key lives in a database row scoped to your tenant id. Cross-customer access is blocked at the data layer. Your prompts and conversations are not visible to anyone but you.

Local-first engine

The Lumen daemon runs on a machine you control. Your conversations and workspace data live there. The hosted layer is the gateway that lets your phone or laptop reach the daemon; the actual work happens on a node you own.

Built for real work

Audit log of every action. Per-tenant API key isolation with AES-256-GCM at rest. RS256-signed 14-day sessions. Workspace memory that survives provider switches. Privacy and Terms in plain language. The serious-software-tool details, not the demo polish.

//The workflow

1. Compose

Open a workspace. Pick a role for the work (or let Coding Mode pick one per turn). Stack the skills that should apply. Choose the AI provider for the task. Your settings persist; you do not re-explain context every conversation.

2. Iterate

Type intent on the left. Lumen drafts a self-contained prompt that carries the expertise lens, the workspace memory, and the constraints. Copy the prompt, run it in your AI agent of choice, paste the result back to refine.

3. Ship

Lumen captures decisions and patterns as you work. When you are ready, render a Blueprint to docs and hand it off. Or just keep iterating; the workspace remembers, the patterns compound.

//Pricing

14-day free trial. No credit card required. Bring your own Anthropic API key.

Monthly

$20/ month

Cancel anytime.

  • All 181 roles + 192 skills
  • Blueprint Designer
  • Unlimited workspaces
  • Workspace memory + retrieval
  • Two-pane workbench
  • Pay with card, Apple Pay, Google Pay, Link, PayPal, Cash App
Start 14-day trial

//Why bring your own key

Lumen charges $20/month for the software. You plug in your own API key from Anthropic, OpenAI, Google, or xAI and pay that provider directly for the model calls you make. That keeps your costs proportional to your actual usage and means heavy users do not subsidize light ones, or the other way around.

//Common questions

Is my ChatGPT Plus / Gemini Advanced subscription enough?

No. Those consumer subscriptions cover the provider's web chat only. Lumen calls the provider's API, which is separately billed. You need an API key from the provider's console. Free-tier API keys work too. Anthropic is the only exception: a Claude.ai Pro / Max subscription works through Claude Code mode.

Why prompts instead of just running the AI directly?

Lumen separates intent-shaping from execution. You stay in control of what runs and on what context; the AI agent on the right pane sees only what you paste. This is the model that produces the best results for serious software work; raw chat with an AI sprawls and forgets, while prompt-by-prompt iteration stays sharp.

Can I switch providers per workspace?

Yes. Each workspace picks its own provider and model independently. You can run a coding workspace on Claude Sonnet 4.6, a research workspace on Gemini 2.5 Pro, and a reasoning workspace on o3, all under one Lumen subscription with your own API keys.

What is Coding Mode?

A workspace setting that auto-picks the best-fit coding role per turn from a curated pool of 25 software-focused roles (coder, tech-lead, devops-engineer, security-auditor, etc.). You type intent, Lumen picks the lens, drafts the prompt. No role-picker fumbling.

What about my data?

Your prompts and chat history live in a tenant-isolated database. No other customer can read your data. The model calls go from us to the provider you chose, using your API key. We do not train any AI on your prompts. See Privacy for the full breakdown.

What if I cancel?

You keep what you built. Your workspaces and prompts remain readable for 30 days after cancellation, then are deleted per our Privacy policy. Cancel anytime from the Billing page; no questions, no retention calls.