Use your own model service
Connect an OpenAI-compatible or Anthropic endpoint while keeping control of its keys and settings.
Chat in the side panel or a full page, then let the agent handle web tasks within clear permission and approval boundaries.
Connect an OpenAI-compatible or Anthropic endpoint while keeping control of its keys and settings.
Reference pages, text, and files, then read, click, type, or download in the browser.
Permission mode, sensitive-site rules, and per-action approval determine what can run.
Panelot is an AI agent extension for Chrome and Edge. You can connect an OpenAI-compatible or Anthropic service, chat in the side panel or full-page interface, and add web pages, text, or files to a task. Browser actions pass through permission policies and approval rules.
Panelot does not provide accounts, cloud sync, or a model proxy. Conversations, settings, approvals, Skills, Plugins, memories, and attachments stay in the current browser profile. Model requests go directly to the provider you configure. When MCP is enabled, tool arguments and relevant results are exchanged with the selected remote server.
| Section | Audience | Contents |
|---|---|---|
| User guide | People installing and using Panelot | Installation, model connections, chat context, browser permissions, Skills, Plugins, MCP, data management, and troubleshooting |
| Development documentation | Contributors and implementation reviewers | Architecture, protocols, storage, providers, the agent engine, tools, security, tests, and releases |
| Privacy policy | All users | Local storage, third-party data flows, retention, and user controls |
chrome://extensions or edge://extensions, enable Developer mode, and choose Load unpacked.See Install and configure for upgrade precautions and the complete procedure.
SKILL.md, install validated data-only Plugins, and connect to remote Streamable HTTP MCP servers.The user guide explains the interface. The development documentation records implementation boundaries and source locations.
Panelot requests website access per origin when needed, and you can revoke it in the browser. Permission mode, user rules, sensitive origins, and the data being sent determine whether an action can run automatically. Users should take over passwords, verification codes, payments, and human-verification steps.
Credentials are encrypted in extension-local storage to reduce plaintext inspection and accidental export. This does not protect against an attacker who can read the browser profile or run code as the current user. See Data and privacy and the privacy policy for data flows and deletion controls.
SKILL.md file. Panelot does not fetch scripts/, references/, or other dependency directories.Development requires Node.js ^20.19.0 || >=22.12.0, pnpm 9.12.3, and Chrome 116 or newer.
pnpm install
pnpm dev
pnpm compile
pnpm lint
pnpm test
pnpm docs:build
pnpm buildSee the development guide for the repository map, verification matrix, and release requirements. If documentation and executable behavior disagree, source code and tests are authoritative.
Panelot is available under the MIT License.