Inject your personal Agent Brain memory into ChatGPT. One memory, every AI.
Adds a 🧠 Brain button below the ChatGPT composer. Click it and the extension will:
- Take your current prompt text as the query
- Recall the top-5 relevant memories from your Brain (
api.agentbrain.ch) - Prepend them silently as a memory-block above your prompt
- ChatGPT sees the memories as context and answers using them
Memories injected this way live only inside the current message — they are not stored in ChatGPT's own memory system.
- Run
chrome://extensions. - Enable Developer mode (toggle top-right).
- Click Load unpacked and select this directory (
agentbrain-chrome-ext/). - Click the Brain icon in the toolbar and hit Settings.
- Paste your API key and workspace ID from https://agentbrain.ch/settings.
- Open https://chatgpt.com/ — a
🧠 Brainbutton appears below the composer.
| Permission | Why |
|---|---|
storage |
Store your API key and workspace ID locally (synced across your Chrome instances) |
activeTab |
Detect when you open ChatGPT |
chatgpt.com, chat.openai.com |
Inject the content script |
api.agentbrain.ch |
Call /memory/recall |
No data leaves your machine other than the recall queries to api.agentbrain.ch using your own API key.
- ChatGPT inject (v0.1)
- Claude.ai inject (v0.2)
- Gemini inject (v0.3)
- Perplexity inject (v0.4)
- After-submit auto-store (optional toggle)
- Similarity-threshold auto-inject (optional toggle)
- Chrome Web Store submission
manifest.json — MV3 config, host permissions
src/background/service-worker.js — single entry for Brain API calls
src/content/chatgpt.js — DOM-inject button + prompt rewrite
src/popup/popup.html|js — toolbar popup status
src/options/options.html|js — settings page
icons/ — (todo: real icons)
Content script ↔ Service Worker ↔ Brain API. Content script never talks to the API directly — avoids CORS headaches and keeps secrets out of the page context.
No build step. Edit files directly and hit the reload icon on chrome://extensions to pick up changes.
MIT