太强了！Hermes Agent 的浏览器扩展来了

结合这篇文章的思路使用，完全告别刷 X 焦虑

现在 Hermes 直接住在每个网页的侧边栏：

• 任意页面实时聊天
• 一键切换 Grok / GPT / Claude / Opus / 本地模型
• 支持视觉分析 + 截图
• 连接本地或远程 Gateway

非官方开源，但社区真爱！浏览器体验直接起飞～

github: https://github.com/abundantbeing/hermes-browser-extension…

abundantbeing/hermes-browser-extension

Source: https://github.com/abundantbeing/hermes-browser-extension

Hermes Browser Extension

Browser-native side panel for Hermes Agent — connect active web context to your local or remote Hermes runtime.

Created by Jon Komet (@abundantbeing). Community extension for Hermes Agent by Nous Research.

Public alpha · Load unpacked · Local/remote Hermes API · Read-only browser context
Not on the Chrome Web Store yet.

What it is

Hermes Browser Extension is not a browser chatbot. It is a Chrome/Edge/Chromium side panel for the real Hermes Agent runtime. It talks to your Hermes Gateway/API server — local by default, remote when you configure a reachable URL — so it can use the models, tools, skills, sessions, memory, and MCP servers already configured in Hermes.

This repo is specifically for the Hermes Browser Extension: the Chrome/Edge/Chromium side-panel integration for Hermes Agent.

Visual tour

Side panel	Theme settings	Local agents

Highlights

• Chrome/Edge/Chromium MV3 side panel powered by the Side Panel API.
• Connects to a configurable local or remote Hermes API server. Default: http://127.0.0.1:8642.
• Supports dashboard WebSocket mode when you have a signed-in remote Hermes dashboard tab and no API key.
• Auto-syncs connected Hermes providers/models, profiles, skills, sessions, and capabilities.
• Shows a Hermes compatibility panel so older gateways degrade into explicit fallback/manual modes instead of broken route errors.
• Sends active tab/browser context into a persisted Hermes session.
• Adds a collapsible “What Hermes saw” receipt after each sent turn for transparent context/debugging.
• Captures active tab title/URL, open tabs, selected text, readable page text, metadata, headings, forms, links, and buttons where available.
• Supports voice dictation through Hermes audio transcription when available, with Browser speech fallback when the connected runtime does not expose STT.
• Wraps webpage text as untrusted context before sending it to Hermes.
• Streams Hermes responses and falls back to non-streaming chat when needed.
• Includes Desktop-style appearance settings: Light/Dark/System mode plus Nous, Midnight, Ember, Mono, Cyberpunk, and Slate themes.
• Includes a localhost agent picker for switching between trusted local Hermes API gateway ports.
• No debugger, nativeMessaging, click/type/form-submit, cookies, history, bookmarks, downloads, or browser-control permissions in v0.1.

Requirements

• Hermes Agent installed and working.
• Hermes Gateway/API server enabled locally or on a reachable remote machine.
• Node.js 20+.
• Chrome, Edge, Brave, Comet, or another Chromium browser with Side Panel API support (Chrome 114+ baseline).

Quick start

1. Clone and build

git clone https://github.com/abundantbeing/hermes-browser-extension.git
cd hermes-browser-extension
npm install
npm run build

The loadable extension is generated at:

dist/

2. Load unpacked in Chrome/Edge

1. Open chrome://extensions or edge://extensions.
2. Enable Developer mode.
3. Click Load unpacked.
4. Select this repo’s dist/ folder — not the repo root and not extension/.
5. Pin/click the Hermes extension icon to open the side panel.

After code updates, run npm run build again and click Reload on the Hermes Browser Extension card in the browser extensions page.

Connect to Hermes

Local API server

Local-only is the safest default. Put this in ~/.hermes/.env on the machine running Hermes:

API_SERVER_ENABLED=true
API_SERVER_HOST=127.0.0.1
API_SERVER_PORT=8642
API_SERVER_KEY=<your-api-server-key>
API_SERVER_CORS_ORIGINS=chrome-extension://<your-extension-id>

Start or restart the gateway:

hermes gateway run

Verify the API server:

HERMES_GATEWAY_URL=http://127.0.0.1:8642
HERMES_API_TOKEN='<your-api-server-key-or-browser-token>'
curl "$HERMES_GATEWAY_URL/health"
curl -H "Authorization: Bearer $HERMES_API_TOKEN" "$HERMES_GATEWAY_URL/v1/models"

Then in the extension side panel:

1. Click Connect to Hermes and approve locally if your Hermes Desktop/gateway supports the approval flow.
2. If approval is not available yet, click Manual setup.
3. Choose Local gateway.
4. Use Gateway URL http://127.0.0.1:8642.
5. Paste your scoped browser token or API_SERVER_KEY.
6. Click Test connection, then Save settings.
7. Open a normal https:// page and ask: Summarize this page in one sentence.

Remote API server

For a remote Hermes machine, bind the API server to a reachable trusted interface and keep CORS narrow:

API_SERVER_ENABLED=true
API_SERVER_HOST=0.0.0.0
API_SERVER_PORT=8642
API_SERVER_KEY=<your-api-server-key>
API_SERVER_CORS_ORIGINS=chrome-extension://<your-extension-id>

Use Tailscale/VPN/reverse proxy + HTTPS where possible. Do not expose the Hermes API server naked to the public internet. The Hermes API server can access the real Hermes runtime and tools.

Remote dashboard mode, no API server

If you run Hermes elsewhere and only expose the OAuth-gated dashboard, select Remote, enter the dashboard’s https:// URL, and leave the API key blank. With no key, the extension connects over the dashboard’s /api/ws socket instead of the REST API server.

Auth uses a single-use WebSocket ticket minted from a signed-in dashboard tab:

• Open the dashboard URL in a normal browser tab and sign in, and keep that tab around.
• The extension mints the ticket first-party from that tab, then opens the socket.
• Test connection opens the socket and loads models, which confirms the whole path.

Limitations in this mode: image attachments are inline-only, and the skills/profiles lists are unavailable because those are REST-only and the dashboard’s REST surface is not reachable cross-origin.

What syncs after connection

After connection, the side panel loads from the connected Hermes gateway:

• /v1/models — all providers/models Hermes can enumerate, including provider-qualified IDs.
• /api/sessions — recent Hermes sessions grouped by source.
• /v1/skills — slash-command skill suggestions in the composer.
• /v1/profiles — profile picker when the gateway exposes profile metadata.
• /v1/capabilities — feature flags such as audio transcription and Browser upload support.

The DOM/context chip should show a non-zero page-context count on normal readable pages. Browser internal pages such as chrome://extensions are intentionally restricted.

Install with Hermes / Computer Use

You can ask Hermes to help install it:

Install Hermes Browser Extension from https://github.com/abundantbeing/hermes-browser-extension. Clone it, run npm install, run npm run build, then use computer use to open chrome://extensions, enable Developer mode, load the dist folder unpacked, and help me connect it to my local or remote Hermes Gateway API server. Do not reveal, print, screenshot, or commit my API key.

Security model

Hermes Browser Extension is intentionally conservative in v0.1:

• Local API server by default; remote API server support requires an explicit URL, token, and CORS allowlist.
• Strong bearer/API key required for API access.
• Page content is wrapped as untrusted context before it reaches Hermes.
• Read-only browser context capture: no click, type, form-submit, checkout, download, or browser-control behavior.
• No debugger, nativeMessaging, cookies, history, downloads, or bookmarks permissions.
• Restricted pages include browser internals, extension pages, and obvious banking/crypto/password/payment/health/government-tax categories.

See SECURITY.md, PERMISSIONS.md, DATA-FLOW.md, and PRIVACY.md for details.

Troubleshooting

I loaded the extension but nothing works

Make sure you loaded dist/, not the repo root. The selected folder must contain manifest.json directly.

The side panel says it cannot connect

Check that Hermes Gateway/API server is running and reachable from the browser:

curl http://127.0.0.1:8642/health
# or, for remote mode:
curl http://<trusted-remote-host>:8642/health

If /v1/models fails, check API_SERVER_KEY, the extension’s stored API key/browser token, and API_SERVER_CORS_ORIGINS. For remote mode, the browser extension origin (chrome-extension://<id>) must be allowlisted on the Hermes machine.

The DOM chip says 0 chars

Open a normal https:// page and refresh context. Browser internal pages (chrome://, edge://, extension pages, devtools, etc.) are restricted by design.

Microphone says blocked or voice dictation does not start

Chromium side panels can suppress microphone permission prompts. Hermes Browser Extension handles this with capability-gated voice modes:

• Hermes STT when the connected Hermes runtime advertises audio transcription.
• Browser speech fallback when Hermes STT is unavailable and Chromium exposes Web Speech.
• A visible Hermes Voice Dictation tab when the side panel cannot capture the mic directly.

Suggested flow:

1. Click the mic button in the side panel.
2. If the side panel cannot capture the mic, a Hermes Voice Dictation tab opens.
3. In that tab, click Start dictation. This click is the permission gesture Chromium expects.
4. Speak, then click Stop + transcribe or Stop speech depending on the active mode.
5. The transcript is sent back to the side panel composer automatically.

If Chromium still says the mic is blocked, click Open microphone settings in the voice tab and set Microphone to Allow for chrome-extension://<the Hermes extension id>/, then return to the voice tab and try again.

The first-run Connect flow is unavailable

Use Manual setup with your local/remote Gateway URL and API key. The native Desktop approval flow is still evolving during alpha.

GitHub PR/Issue auto-review

This repo includes two Hermes review runners:

• npm run review:watch — local poller for open PRs/issues. This works now from a machine that can reach Hermes and is authenticated with gh.
• npm run review:event — GitHub-event runner for future GitHub Actions/webhook wiring. It expects GITHUB_EVENT_NAME, GITHUB_EVENT_PATH, and GITHUB_REPOSITORY.

The local poller checks open PRs and issues, computes a stable signature from PR head SHA or issue title/body, and only reviews changed targets. It upserts one bot comment per PR/issue with a stable marker. PR diffs and issue bodies are treated as untrusted input.

Local setup:

# Uses gh auth token, local API_SERVER_KEY from ~/.hermes/.env,
# and http://127.0.0.1:8642 by default.
npm run review:watch

Optional overrides:

HERMES_REVIEW_REPO=abundantbeing/hermes-browser-extension
HERMES_REVIEW_GATEWAY_URL=http://127.0.0.1:8642
HERMES_REVIEW_API_KEY=<api-server-key-or-scoped-token>
HERMES_REVIEW_MAX_TARGETS=3
HERMES_REVIEW_STATE_FILE=~/.hermes/hermes-browser-review-state.json

For a GitHub-hosted Actions runner later, HERMES_REVIEW_GATEWAY_URL must be reachable from GitHub. A runner cannot reach http://127.0.0.1:8642 on your personal machine; use a remote Hermes API server behind Tailscale/VPN/HTTPS or a self-hosted GitHub runner on the same network. Pushing .github/workflows/* also requires a GitHub token with workflow scope.

Dry-runs:

npm run review:watch:dry-run

GITHUB_EVENT_NAME=pull_request_target \
GITHUB_EVENT_PATH=./event.json \
GITHUB_REPOSITORY=abundantbeing/hermes-browser-extension \
GITHUB_TOKEN=<github-token> \
npm run review:event:dry-run

Development

npm test
npm run check:js
npm run check:manifest
npm run verify
npm run build
npm run package

Project layout:

extension/
  manifest.json       MV3 extension manifest
  background.js       side panel behavior
  content.js          page context collector
  sidepanel.html      side panel UI
  sidepanel.css       side panel styling
  sidepanel.js        Hermes API client + UI state
  voice-dictation.*   visible extension voice recorder fallback for blocked side-panel mic capture
  request-permissions.* visible extension mic-permission helper page
  sidepanel-preview.html static visual QA preview
  assets/             local Hermes fonts, icons, and imagery
  lib/common.mjs      shared prompt/context/security utilities
scripts/
  build.mjs           copies extension/ to dist/
  check-manifest.mjs  validates required manifest assets/permissions
  hermes-review-github-event.mjs PR/issue event runner for GitHub Actions/webhooks
  hermes-review-watch.mjs local PR/issue review poller
  package.mjs         creates artifacts/hermes-browser-extension.tar.gz
tests/
  common.test.mjs     utility behavior tests

Relationship to Hermes Agent

Hermes Agent is an open-source project by Nous Research. Hermes Browser Extension is a community extension by Jon Komet that connects to a local or remote Hermes API server. It is designed to live at the edge of the ecosystem without adding core tool-schema footprint.

Useful links:

• Hermes docs: https://hermes-agent.nousresearch.com/docs
• Hermes API server docs: https://hermes-agent.nousresearch.com/docs/user-guide/features/api-server
• Hermes upstream repo: https://github.com/NousResearch/hermes-agent

Author

Built by Jon Komet (@abundantbeing).

License

MIT. See LICENSE.