Understand the workspace
Read the project, files, and relevant memories before deciding what the task really needs.
Context alignedAn AI work team that runs on your computer
Turn one request into
work that is actually done
Lume goes beyond answers. It understands your workspace, uses local tools, remembers long-term preferences, and sends a specialist agent team to deliver usable results.
MIT open sourceLocal dataYour choice of model
Lume workspace
Prompt
“Research this market and turn the findings into a presentation-ready brief.”
Not a smarter chat box.
A workbench that understands, delegates, and acts.
One task, end to end
You set the goal. Lume runs the process.
From context to delivery, every step stays visible, pausable, and resumable.
Plan and delegate
Break a complex goal into clear pieces and assign each one to the best-fit specialist.
Plan stays editableUse tools and act
Search, edit files, run commands, and generate images instead of stopping at advice.
Progress stays visibleDeliver usable work
Return conclusions and artifacts in a thread you can keep moving forward.
Results saved locallyFrom idea to artifact
Work, create, and research from one place
No elaborate prompt craft required. Describe the real need and Lume finds the next step.
Research
Turn scattered sources into a reliable point of view
Search across sources, verify claims, map disagreements, and deliver a cited brief.
L
“Compare five competitors and show me where the real opportunity is.”
3 agents are working in parallel
Build
Move real work forward inside a codebase
Understand constraints, locate files, edit code, and verify the result with full context intact.
L
“Fix this issue without changing the existing sign-in flow.”
3 agents are working in parallel
Create
Go from a blank page to something worth sharing
Research, writing, translation, and visuals work together to shape a complete piece.
L
“Turn these interviews into an opinionated long-form article.”
3 agents are working in parallel
Automate
Let recurring work happen on schedule
Create routines and automations, then receive the output in your messaging channel.
L
“Every Monday, summarize progress and send me the top risks.”
3 agents are working in parallel
Context that stays
It remembers more than chat history
Project context, writing preferences, key decisions, and unfinished work carry into the next task. Memory is stored as readable local files you can inspect, edit, back up, or move.
Preference: lead with the conclusion
Constraint: this project uses Bun
In progress: website redesign, mobile review pending
“No more starting from zero. Lume knows what you are doing and how you like to work.”
Local-first by design
Your work does not need to be uploaded first
Memories, threads, skills, automations, and agent workspaces live under ~/.lume/. Markdown and YAML remain the source of truth.
~/.lume/
├── memory/ # Markdown
├── skills/ # SKILL.md
├── automation/ # cron
├── agent-workspaces/
└── lume.yaml- Readable: no proprietary data lock-in
- Controlled: you approve sensitive actions
- Portable: copy one folder to back up or move
Built to act
One workbench, a complete execution layer
Files, terminals, browsers, search, image generation, Office artifacts, and MCP services all share the same task context.
File systemTerminalWeb & searchImage generationMCP servicesMessagingAutomationDocuments
Work continues when you step away
Quietly running. Ready on time.
Hand recurring work to automations. Lume runs locally on schedule, brings in the right personas and tools, then sends the result to the channels you already use.
1 Scheduled research and reports
2 Parallel specialist execution
3 Proactive result delivery
Next task · Today at 10:00 PM
陆寻 Explorer 沈策 Planner 审岳 Code Reviewer 顾砚 Researcher 许澄 Translator 江岚 Writer 宋澈 Voice Artist 
林澄 Design Engineer 白洛 Illustrator 唐栩 Data Analyst 
纪衡 Quant Analyst 
温序 Novelist 祁远 Developer 陆寻 Explorer 沈策 Planner 审岳 Code Reviewer 顾砚 Researcher 许澄 Translator 江岚 Writer 宋澈 Voice Artist 林澄 Design Engineer 白洛 Illustrator 唐栩 Data Analyst 纪衡 Quant Analyst 温序 Novelist 祁远 Developer Not one generic persona
A team with distinct specialties
Every persona has a clear specialty, workspace, and way of working. The main thread understands your goal and hands each part to the right person.
Personas are functional: each has its own system prompt, tool preferences, and task boundaries.
陆寻
Explorer
沈策
Planner
审岳
Code Reviewer
顾砚
Researcher
许澄
Translator
江岚
Writer
宋澈
Voice Artist
林澄
Design Engineer
白洛
Illustrator
唐栩
Data Analyst
纪衡
Quant Analyst
温序
Novelist
祁远
Developer
13 specialists, ready when needed
Keep your model choice
One interface, the models you trust
Connect leading providers and local models. Assign models by task and balance quality, speed, and cost on your terms.
OpenAIAnthropicGeminiDeepSeekQwenKimiGLMDoubaoOpenRouterOllamaLM Studio
Made for builders too
Put the same agent engine inside your product
Lume ships its core loop as a standalone SDK. Run it in a server, Docker, serverless, or CI without a local CLI dependency.
Read the SDK docsagent.ts
import { createAgent } from '@lume/agent-sdk'
const agent = createAgent({
model: 'claude-sonnet-4-6',
maxTurns: 10,
})
for await (const event of agent.query(
'Read the project and ship the change.'
)) {
console.log(event)
}FAQ
A few things to know before you start
How is Lume different from a normal AI chat app?+
Chat apps optimize for one-off answers. Lume is built for ongoing work: it can use local files and tools, keep durable memory, and delegate complex tasks across specialist personas.
Where is my data stored?+
Core data lives in the ~/.lume/ folder on your computer. Memory uses Markdown as its source of truth, with readable configuration formats for easy backup and migration.
Can I use my own API key or a local model?+
Yes. Lume supports OpenAI-compatible endpoints, a range of hosted providers, and local options such as Ollama and LM Studio.
Is Lume free?+
Lume is free to download and open source under the MIT license. Third-party model providers may charge according to their own pricing.
Bring AI into the actual workspace
Give the next task to Lume
Local first, free, and open source. Get started in minutes.