Lumo for Chrome
Lumo
An agent that acts inside the page.
A sidebar agent that plans, calls browser and DevTools tools, checks the result, and keeps going until the task is done.
Plan. Call tools. Check. Repeat.
One turn can run as many steps as the task needs, with the whole tool chain visible and a stop button that works.
- As many steps as the task needs
- Parallel tool calls with typed inputs
- Stop, retry or continue without losing state
One turn, many steps
model ↔ tools until the goal is met
Goal + page context
Prompt, selection, tab and attachments
Model step
Reasoning, plan, then a tool call
Tool execution
MCP calls run in parallel
Result + next step
The loop decides again
Browser tools, plus DevTools.
Built-in browser tools, remote MCP, WebMCP sites and other extensions share one tool registry.
Browser Core
Tabs, windows, navigation, bookmarks, history, cookies, downloads.
Page Interaction
Page reading, snapshots, clicks, forms, keyboard, screenshots.
Network Monitor
Request capture, blocking, redirects and header rewriting.
DevTools Advanced
CDP sessions, trusted input, emulation and console capture.
File Manager
Read, write, patch and preview files stored locally.
External MCP
HTTP Streamable or SSE, with visible status and errors.
Extension MCP
Sibling extensions serve MCP over chrome.runtime.Port.
WebMCP
Pages declare their own tools for the active tab.
The layer a content script cannot reach.
Trusted input, device and network emulation, console capture — through CDP.
chrome.debugger · CDP 1.3debug_sessionAttach or detach the debugger.
debug_inputClick, type, press keys and drag.
debug_emulateDevice, network, geolocation and CPU.
debug_consoleConsole messages and uncaught errors.
One conversation, or three.
Run independent conversations side by side. Each pane keeps its own model, history and tool calls; drag the dividers to resize.
Your model. Your key. Three API types.
Lumo has no model of its own. Point it at Chat Completions, Responses or Anthropic Messages.
- Chat Completions, Responses and Anthropic Messages
- Vision, reasoning effort and context window per model
- Keys stay local; requests go straight to your provider
Three API types
One picker, one key, any compatible endpoint.
Chat Completions
The portable default for most gateways.
POST /v1/chat/completionsResponses
OpenAI's newer endpoint, with reasoning.
POST /v1/responsesAnthropic Messages
Claude models, vision and long context.
POST /v1/messagesYour data stays on your machine.
Conversations, files and settings live in extension storage. No account, no relay.
- No account, no telemetry
- Files stay in IndexedDB
- Every permission is documented
Built for the real web.
Quick actions
Send the page or selection from the right-click menu.
Attachments
Page, selection, image, file or pasted text.
OCR
Read images with a non-vision model.
Local files
The agent writes and previews files in IndexedDB.
Conversation history
Search and reopen past turns locally.
Chat debug
Inspect the prompt and tool calls behind a reply.
Three ways in.
Store for stability, an Actions build for the newest fixes, or source.
Also from Lumo
Need the page rewritten?
Lumo Userscripts exposes its scripts over MCP, so the agent can write and manage them with you.