Carry your AI conversation to any platform — without losing context.
Continue it is a zero-dependency Chrome extension (Manifest V3) that extracts a full conversation from one AI chat platform and packages it into a portable prompt you can drop into any other platform — instantly resuming where you left off.
When you're deep in a conversation with Claude, ChatGPT, Gemini, Grok, or Perplexity and want to switch platforms (or start a fresh session while keeping context), Continue it:
- Scrolls through and captures your entire conversation history from the page DOM
- Generates a summary of the thread — task, requirements, decisions, blockers, files, timeline. Choose one of three modes: local (no AI), shared Server AI (5 free/day), or your own API key (unlimited) — see AI summaries
- Packages the context into a prompt (or a sequence of chunks for long conversations) ready to paste into any target LLM
- Tracks chunk progress in the popup so you can send large transcripts in staged batches
| Platform | Extract from | Import to |
|---|---|---|
| Claude | claude.ai | ✓ |
| ChatGPT | chatgpt.com, chat.openai.com | ✓ |
| Gemini | gemini.google.com | ✓ |
| Grok | grok.com, x.com/i/grok | ✓ |
| Perplexity | perplexity.ai | ✓ |
- Auto-scrolls up to 80 steps (700 px each) to load the full chat history
- Role detection via DOM attributes, ARIA labels, and keyword heuristics (7-level parent scan)
- Deduplication by content hash — no repeated messages even with dynamic loading
- Noise filtering strips buttons, UI labels, and short non-message text
- Stability detection stops early when no new messages appear across multiple scroll steps
Three verbosity levels — short, medium (default), detailed — each producing:
| Section | What it captures |
|---|---|
| Task | Primary objective + current request |
| Timeline | Sampled messages across the conversation arc |
| Key requirements | Scored user-intent sentences |
| Assistant findings | Scored assistant insights and outputs |
| File references | Paths, filenames, CSS selectors, backtick tokens |
| Blockers | Error and problem statements |
| Decisions | Choices discussed and confirmed |
| Next action | Recommended continuation prompt |
Sentences are scored by length, position, message order, and keyword relevance before selection. This local summary is always produced as a baseline; the AI modes below refine it into a higher-quality handoff.
Direct import (≤ 6,500 tokens) Single prompt with header + summary + chunk digest + import instructions. Paste once, continue.
Staged import (> 6,500 tokens)
- A starter prompt the target AI acknowledges first
- Transcript divided into ~12 KB chunks sent one at a time
- The popup's Copy next chunk button advances the cursor automatically
- Stats: source platform, captured-at timestamp, message counts, chunk progress
- Summary mode selector (persisted to storage)
- Editable summary and prompt previews
- Buttons: copy recommended prompt, copy next chunk, reset chunk queue, download JSON, clear handoff
| Key | Contents |
|---|---|
continueIt.latestHandoff |
Full handoff object (JSON) |
continueIt.summaryMode |
Preferred verbosity level |
continueIt.chunkCursor |
Per-handoff chunk index for staged imports |
continueIt.handoffHistory |
Last 10 exports (metadata only) |
continueIt.ai.mode |
AI summary mode: none / server / byok |
continueIt.ai.serverUrl |
Shared backend URL (Server AI mode) |
continueIt.ai.byokBaseUrl / byokModel / byokApiKey / byokProvider |
Your own provider config |
continueIt.clientId |
Anonymous id used for the Server AI daily quota |
Handoff data lives in chrome.storage.local. In No AI mode nothing ever leaves the browser. In Server AI / Your own key modes, a compacted summary of the conversation is sent to the endpoint you choose (see below).
Pick a mode in the popup under AI summary mode. All three produce the same portable handoff; they differ only in how the summary is written.
| Mode | Quality | Cost | Privacy | Setup |
|---|---|---|---|---|
| No AI (default) | Good (local heuristic) | Free | Fully local, offline | None |
| Server AI | Better (real LLM) | Free, 5 exports / 24h | Summary sent to the shared backend | None for the user |
| Your own API key | Best (any model you like) | Free on the providers below | Summary sent to your chosen provider | Paste a key |
If an AI request ever fails (rate limit, bad key, network), the extension automatically falls back to the local summary and adds a warning — an export never breaks.
Any OpenAI-compatible endpoint works. Create a key (most need no credit card), pick it in the popup, paste the key, and click Save AI settings (this also grants the extension permission to reach that host). Use Test connection to verify.
| Provider | Base URL | Example free model | Get a key |
|---|---|---|---|
| OpenRouter | https://openrouter.ai/api/v1 |
meta-llama/llama-3.3-70b-instruct:free |
openrouter.ai/keys |
| Google AI Studio | https://generativelanguage.googleapis.com/v1beta/openai |
gemini-2.0-flash |
aistudio.google.com/apikey |
| Groq (fastest) | https://api.groq.com/openai/v1 |
llama-3.3-70b-versatile |
console.groq.com/keys |
| Cerebras (highest volume) | https://api.cerebras.ai/v1 |
llama-3.3-70b |
cloud.cerebras.ai |
| NVIDIA NIM | https://integrate.api.nvidia.com/v1 |
meta/llama-3.1-70b-instruct |
build.nvidia.com |
| Mistral | https://api.mistral.ai/v1 |
mistral-small-latest |
console.mistral.ai |
Free-tier limits change often — verify the current numbers on each provider's site. Keys are stored only in chrome.storage.local on your machine and are sent only to the provider you configured.
The Server AI mode lets you offer smart summaries to users without them needing a key, capped at 5 exports per 24h per user (by anonymous client id + IP) so it stays cheap.
cd server # from the repo root
npm install # installs express, cors, dotenv (once, at repo root)
cp .env.example .env # then edit .env
npm startConfigure .env with any free OpenAI-compatible provider so it costs you nothing:
OPENAI_BASE_URL=https://openrouter.ai/api/v1
OPENAI_API_KEY=sk-or-... # your key, never exposed to users
OPENAI_MODEL=meta-llama/llama-3.3-70b-instruct:free
DAILY_LIMIT=5 # exports per window
RATE_WINDOW_HOURS=24Then set the Server URL in the popup (default http://localhost:8787) and choose Server AI. Deploy the server anywhere (Render, Railway, Fly, a VPS) and point the popup at that URL.
The in-memory rate limiter is per-instance and resets on restart. For a hardened multi-instance deployment, back it with Redis or a database.
Continue it is an unpacked extension — no Chrome Web Store listing yet.
- Download or clone this repository
- Open Chrome and go to
chrome://extensions/ - Enable Developer mode (toggle in the top-right corner)
- Click Load unpacked or Load packed if you have a zip file.
- Select the
Continue itfolder - The extension icon appears in your toolbar
Supports any Chromium-based browser that handles Manifest V3: Chrome, Edge, Brave, Arc.
- Open any supported AI chat page with an active conversation
- Click the Export Context button that appears on the right side of the page
- The extension scrolls through and captures the full thread
- A modal opens showing:
- Conversation stats (messages, roles, estimated tokens, chunks)
- Editable summary (switch mode to regenerate)
- Recommended prompt or chunk-by-chunk controls
- Click Copy recommended prompt (or download the raw JSON)
Short conversations (direct mode):
- Paste the copied prompt into the target AI's chat input and send
Long conversations (staged mode):
- Paste the starter prompt and send — the AI acknowledges the incoming context
- Open the extension popup
- Click Copy next chunk → paste → send
- Repeat until all chunks are sent
- The target AI now has the full conversation history reconstructed
| Button | Action |
|---|---|
| Copy recommended prompt | Copies the smart-selected prompt (direct or starter) |
| Copy next chunk | Copies next transcript chunk and advances the cursor |
| Reset chunk queue | Resets cursor to chunk 0 |
| Download JSON | Downloads the full handoff as a .json file |
| Clear saved handoff | Wipes the current handoff from storage |
Continue it/
├── manifest.json # Extension config (MV3), permissions, host permissions
├── background.js # Service worker — performs AI summarize requests (Server AI + BYO)
├── provider-config.js # Provider registry (selectors, role hints per platform)
├── shared-handoff.js # Core data model, local summarizer, chunker, storage API
├── shared-ai.js # Unified AI client — modes, provider presets, settings, permissions
├── shared-ui.js # Toast, modal, and launcher button components
├── content-site.js # Generic content script injected on all supported sites
├── popup.html/.css/.js # Popup UI (stats, AI mode selector, handoff controls)
└── server/
└── server.js # Optional shared backend — OpenAI-compatible upstream + rate limiter
provider-config.js → shared-handoff.js → shared-ai.js → shared-ui.js → content-site.js
(provider registry) (data model/API) (AI client) (UI library) (button + modal)
The content script never calls an AI provider directly (page CSP/CORS would block it on strict sites). Instead it hands a compacted summary to the background service worker, which holds the host permission and makes the cross-origin request:
// content-site.js → shared-ai.js → background.js
chrome.runtime.sendMessage({ type: "continueIt.summarize", payload }, cb)
// → Server AI: POST <serverUrl>/api/summarize (with x-continue-it-client header)
// → Your key: POST <baseUrl>/chat/completions (Authorization: Bearer <key>)
// Response shape
{ ok, used, summary, quota?: { limit, remaining, resetAt }, error }{
schemaVersion: 2,
id: "handoff_<timestamp>_<hash>",
source: "Claude" | "ChatGPT" | "Gemini" | "Grok" | "Perplexity",
createdAt: "<ISO 8601>",
pageTitle: string,
pageUrl: string,
summaryMode: "short" | "medium" | "detailed",
summary: string,
messages: [{ id, role, text }],
stats: {
totalMessages, userMessages, assistantMessages,
unknownMessages, totalCharacters, estimatedTranscriptTokens
},
diagnostics: { hitStepLimit, scanSteps },
warnings: string[],
detectedFiles: string[]
}| Limit | Value |
|---|---|
| Max scroll steps per extraction | 80 |
| Scroll step size | 700 px |
| Direct import token threshold | 6,500 tokens |
| Transcript chunk size | ~12,000 characters |
| Storage warning threshold | 4 MB |
| Handoff history kept | 10 entries |
No build step, no dependencies.
# Edit any file directly, then reload the extension:
# chrome://extensions/ → Continue it → ↺ (reload icon)The extension is plain ES6+ JavaScript — no TypeScript, no bundler, no npm.
To test on a platform, navigate to a chat page with an active conversation and trigger the Export Context button.
| Permission | Why |
|---|---|
storage |
Persist handoffs and chunk cursors across sessions |
unlimitedStorage |
Allow large conversation exports without hitting the default 5 MB quota |
clipboardWrite |
Copy prompts and chunks to clipboard |
| Host permissions (8 domains) | Inject content scripts on supported AI platforms |
optional_host_permissions |
Requested only when you enable Server AI or your own API key — grants access to that one endpoint |
In No AI mode, no network requests are made and no data leaves your browser. In Server AI or Your own API key mode, a compacted summary is sent only to the endpoint you configured, and the extension asks for permission to reach that host at the moment you save the setting.
MIT © 2026 Rohit Nehte
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
0.7.0 — Manifest V3, schema v2. Adds three AI summary modes (local / Server AI with 5-per-day quota / bring-your-own OpenAI-compatible key) and an optional rate-limited backend.