Research
Research & briefing
Read across a dozen open tabs and come back with one structured brief instead of twelve summaries.
Replaces: an afternoon of copy-pasting into a doc.
Chrome extension · Free · Bring your own key
AI agents that work where you already work.
InnerAgents lives in your Chrome sidebar. It reads the page you're on, clicks through the apps you use, and connects to your tools — so the repetitive part of your day stops being your problem.
No account. Your own API key. Your conversations go straight to the model provider you picked — and to nobody else.
Bring your own model — and change your mind later.
- Anthropic Claude
- OpenAI
- Google Gemini
- Mistral
- OpenRouter
- Ollama
- Model Context Protocol
How it works
Three steps, then it's just part of your browser.
-
1
Connect your model
Paste an API key from Anthropic, OpenAI or Mistral — or point it at a model running on your own machine.
-
2
Give it context and tools
One click sends the current page, a screenshot or a selection to your agent. Connect your calendar, your knowledge base, your ticketing system.
-
3
Let it do the work
It fills forms, navigates apps, extracts data — and asks you to confirm before anything irreversible.
Use cases
For the work that fills the day but never finishes it.
Data
Getting data out of apps
Pull records out of a CRM, an ATS or an internal back-office, then review and correct them in an editable form before they're used anywhere.
Replaces: an export feature that doesn't exist.
Operations
Filling things in
The agent works the interface you'd work: typing in fields, choosing from dropdowns, moving from screen to screen.
Replaces: the same twelve clicks, forty times.
Integration
Your tools, in the conversation
Connect calendars, databases and internal APIs through MCP servers. The agent uses them the way it uses the browser.
Replaces: switching between six tabs to answer one question.
What it can do
An agent with hands, not just opinions.
Sees your screen
Page content, a screenshot, or just the text you selected.
Acts in the browser
Clicks, types, navigates, and waits for the page to load.
You approve what matters
A review form appears before extracted data is used. Destructive actions always ask first.
Remembers you
Facts kept between conversations, stored on your machine.
Several agents, one job
A research agent hands off to a writing agent. The whole chain stays visible.
See what it costs
Token count and estimated price per call, before a bill surprises you.
Security & privacy
By default, nothing leaves your machine.
Most AI tools route your work through their infrastructure. This one keeps everything local by default. Conversations, keys and memory stay in your browser's local storage; requests go straight from your machine to the model provider you chose.
Read the privacy policyBring your own key
No InnerAgents account, no shared quota, no intermediary billing.
Anonymous telemetry, opt-out
Usage counts and crash reports help fix bugs — never message content, page content, keys or URLs. One toggle in Settings stops all of it.
Credential vault
Keys and tokens are stored once and referenced by name. The model sees the name, never the secret.
Permission by permission
Every browser capability an agent uses is granted explicitly by you, the first time it's used.
Confirmation before damage
Deleting anything asks first. Every time.
Local models supported
Point InnerAgents at a model running on your own hardware and nothing leaves the machine at all.
Independence
Your agent layer shouldn't belong to your model vendor.
InnerAgents is tied to no AI company. The agents, their tools, their memory and the connections to your systems live in your browser. The model underneath is a setting.
Change model, keep everything else
Switch from one provider to another and the agent, its tools and its memory carry over untouched. Run two in parallel if you're comparing.
Your contract, your rates
You buy your tokens directly from the provider you negotiated with. Nothing is resold to you, and no third party sits between your team and its model.
Standard across a team
Agents are files. Review one, hand it to everybody, keep the configuration under your own control — no console required to get started.
Marketplace
Start from an agent that already works.
Import a ready-made agent in one click, or export yours and share it with your team as a single file.
Browse the marketplacePage summariserTurns any long page into five bullets.
Table extractorScrapes records into a reviewable table.
Inbox triageSorts and drafts, never sends.
Form fillerCarries data from one app to another.
FAQ
Including what it doesn't do.
Do I need an account?
No. Install the extension, add your API key, done. There is nothing to sign up for.
What does it cost?
The extension is free. You pay your model provider directly, at their prices. The extension shows you the estimated cost of each call.
Which browsers does it work in?
Chrome and Chromium-based browsers. Firefox and Safari don't offer the sidebar API it's built on.
What exactly do you collect?
Anonymous usage counts (the extension was opened, a message was sent) and crash reports, to find bugs. Never the content of your messages, the pages you visit, your keys or your URLs. The identifier is a random UUID created at install. Settings → General → Telemetry turns it off immediately, for analytics and crash reports alike.
Are there pages it can't read?
Yes. Browser system pages, the Chrome Web Store, PDFs opened natively by Chrome, and some sites with a strict security policy.
Can I use a model that runs on my own machine?
Yes, through any OpenAI-compatible endpoint — Ollama and LM Studio included.
Can it work on a schedule, without me?
Not today. An agent runs when you're there and the sidebar is open.
Can my team share agents?
Yes. Export an agent as a single file; anyone can import it, with conflicts handled at import time.
Try it on the tab you have open right now.
Add to ChromeFree · No account · Works with the model you already pay for.