AI Chat in Side Panel
Conversational interface that lives alongside your browsing, no tab switching, no context loss.
Open source · MIT licensed · Manifest V3
Your AI agent,
right in your Chrome side panel.
Gecko Agent reads web pages, clicks, types, and runs multi-step workflows for you. Connect any LLM via OpenRouter and let it think, act, and observe, until the task is done.
Free and open source. Bring your own OpenRouter API key.
Features
Everything you need to automate the browser.
Eleven capabilities, zero fluff. Built for sales pros, recruiters, marketers, and anyone tired of repetitive web tasks.
Multi-Model Support
Pick Claude, GPT, Gemini, Kimi, or any model available on OpenRouter. Bring your own API key.
9 Browser Automation Tools
Click, type, scroll, navigate, read pages, fill inputs, and more, all controlled by the agent.
Agentic Loop
Configurable up to 40 iterations of think → act → observe for complex multi-step tasks.
Real-Time Streaming
Token-by-token response display. See the agent think and act live, no spinner-and-wait UX.
LinkedIn Optimization
Structured data extraction from LinkedIn profiles via JSON-LD. Perfect for sourcing and outreach.
Airtable Support
Smart text-based clicking handles Airtable's dynamic CSS classes that break traditional selectors.
Activity Log
Track every tool execution with status, target, and timestamp, full transparency on what the agent did.
Conversation History
Save and restore past conversations. Pick up exactly where you left off, any time.
Custom Prompts
Create and manage reusable prompt templates. Build your own library of agent recipes.
Guided Onboarding
Step-by-step first-launch setup. New users are productive within 60 seconds, no docs needed.
How it works
Think. Act. Observe. Repeat.
Gecko Agent runs a continuous reasoning loop, up to 40 iterations, until your task is complete.
STEP 01
Think
The AI analyzes your request and the current page state. It decides which tool to call next based on what it sees.
STEP 02
Act
It executes a browser action, click a button, fill an input, navigate to a URL, scroll, or extract data.
STEP 03
Observe
It reads the result of the action, updates its understanding of the page, and decides what to do next. Loops until done.
Hit the Stop button anytime to cancel mid-execution.
Available tools
9 atomic browser actions, infinite combinations.
The agent picks the right tool for each step and chains them until your task is done.
| Tool | Description | Example prompt |
|---|---|---|
read_page
|
Extracts the page content (HTML or structured text). Smart JSON-LD parsing on LinkedIn. |
Read this page and summarize it
|
goto_url
|
Navigates the current tab to a target URL. |
Go to https://example.com
|
open_tab
|
Opens a new browser tab with a given URL. |
Open LinkedIn in a new tab
|
click_selector
|
Clicks an element by CSS selector. Supports double-click. |
Click the submit button
|
click_text
|
Finds and clicks an element by its visible text. Position control: on / right / below. |
Click the cell next to 'Company Name'
|
fill_input
|
Fills a text input, textarea, or contenteditable element. |
Fill the email field with john@example.com
|
type_text
|
Types text into the currently focused element, character by character. |
Type 'Hello' into the active cell
|
scroll_page
|
Scrolls the page by a pixel amount or to a specific element. |
Scroll down to see more results
|
press_key
|
Presses a keyboard key (Tab, Enter, Escape, arrows…) with optional modifiers. |
Press Enter to submit
|
Airtable users:
click_text and
type_text
are built for Airtable's dynamic CSS classes, they just work.
Models
Pick your favorite LLM. Switch anytime.
Gecko Agent connects via OpenRouter, one API key, every major model. Bring your own key, pay only what you use.
Claude 4.8 Opus
Anthropic
Complex reasoning, code generation
GPT-5.5
OpenAI
Agentic tasks
Gemini 3.5 Flash
Fast responses, simple tasks
Kimi K2.6
Moonshot AI
Agentic tasks, open-weight
+ Any of the 300+ models on OpenRouter
Mistral, DeepSeek, Qwen, Command R, and many more. Switch in two clicks from the Settings tab.
FAQ
Questions, answered.
Still on the fence? Here are the eight things people ask us most.
Is Gecko Agent free?
Yes, the extension is 100% free and open source (MIT license). The only cost is your LLM usage on OpenRouter, pay-as-you-go, typically a few cents per task. You bring your own API key, we never see your traffic.
Do I need to know how to code?
No. You talk to the agent in plain English (or any language). It picks the right browser actions for you. The only one-time setup is installing the extension and pasting your OpenRouter API key, guided onboarding handles both in under 60 seconds.
Which LLM models can I use?
Claude (Opus, Sonnet, Haiku), GPT (5, 4o, 4 Turbo), Gemini (Pro, Flash), Kimi (K2, K2.6), Mistral, DeepSeek, Qwen, and 300+ other models available on OpenRouter. Switch between them in two clicks from the Settings tab.
Is my data safe?
Yes. Your API key, conversation history, and prompts are stored locally in Chrome's secure storage, never sent to our servers (we don't have servers handling your data). The agent only sends what's needed to OpenRouter when you trigger a task. Read our full Privacy Policy for details.
Why OpenRouter and not a built-in API key?
Because you stay in control. You see exactly what you spend, you can switch providers anytime, and we never become a middleman who can throttle you or change pricing. Your key, your bill, your data.
Does it really work on LinkedIn and Airtable?
Yes, these are first-class supported use cases. LinkedIn has dedicated JSON-LD extraction for clean profile data, and Airtable benefits from the click_text and type_text tools that bypass dynamic CSS classes. Other complex web apps (Notion, Salesforce, Gmail) also work, but may need a more detailed prompt.
Can the agent break things or click on something dangerous?
The agent only runs when you trigger a task, and you can hit the Stop button at any time mid-execution. It cannot install software, change browser settings, or access tabs you haven't authorized. For destructive actions (delete, submit payments), we recommend supervising the first few runs and adjusting your prompts.
Where can I report bugs or request features?
Open an issue on our GitHub repository. We read everything and ship fixes regularly. Pull requests are welcome too, the codebase follows strict TypeScript + clean architecture conventions documented in DEV-RULES.md.
Available now on Chrome Web Store
Install Gecko Agent in two clicks.
Free and open source. MIT licensed. Bring your own OpenRouter key and start automating in under 60 seconds.
No credit card, no signup. Works on Chrome, Edge, Brave and any Chromium-based browser.