CLAMP — Context Lifecycle & Memory Protocol

Stop burning your AI session limits.
One click. Conversation compressed. Fresh session opened. Ready to continue.

---

The problem nobody talks about

You're deep in a coding session with Claude or ChatGPT. The AI knows your entire codebase, your decisions, your half-finished architecture. The context is rich.

Then the session fills up. You open a new chat.

And spend the next 20 minutes re-explaining everything from scratch — burning half your new session limit before writing a single line.

This happens to every serious AI user. Every day.

The root problem isn't context length. It's that a 60,000-token conversation has maybe 4,000 tokens of load-bearing signal. The rest is noise — repeated file pastes, error traces, scaffolding that was useful once.

What CLAMP does

CLAMP is a browser extension. One click while you're in any AI chat:

1. Reads your entire conversation
2. Distills it into a structured CLAMP.md — task, stack, decisions, blockers, resume
3. Opens a new chat on the same platform
4. Pastes the summary into the input box, ending with Your prompt: so you continue immediately

Zero re-explaining. Zero context lost. The session is fresh but the AI already knows everything.

Install

Option A — Load unpacked (30 seconds)

1. Download clamp-v0.2.0.zip and unzip
2. Go to chrome://extensions
3. Enable Developer mode (top right toggle)
4. Click Load unpacked → select the clamp-ext folder
5. Click the CLAMP icon → Settings → paste your OpenRouter API key → Save

Option B — Chrome Web Store
(coming soon — star this repo to get notified)

Supported platforms

Platform	Status
claude.ai	✅ Supported
chatgpt.com	✅ Supported
gemini.google.com	❌ Not pasting new prompt in new chat, working on it
grok.x.com	🔜 Coming soon
perplexity.ai	🔜 Coming soon

What a CLAMP.md looks like

# CLAMP v1

## TASK
Building a Chrome extension that compresses AI chat sessions and continues
them in a fresh tab. Currently working on the DOM injection into new tabs.

## STACK
- Manifest V3, vanilla JS (no build step)
- OpenRouter API for distillation
- Supports Claude, ChatGPT, Gemini

## STATE
Done: scraping, distillation API, popup UI, new tab opening
Active: fixing injection timing on ChatGPT's SPA
Pending: Firefox port, Chrome Web Store submission

## DECISIONS
1. executeScript over content scripts — works on pre-existing tabs, no install reload needed
2. OpenRouter over direct APIs — model-agnostic, user brings their own key
3. No build step — zero friction for contributors, load unpacked just works

## BLOCKERS
ChatGPT occasionally hydrates the input box late — retry loop handles it but
timing needs tuning on slow connections.

## RESUME
You are helping build CLAMP, a Chrome extension (Manifest V3) that scrapes
AI chat conversations, distills them via OpenRouter, and injects the summary
into a new chat tab. The scraping and distillation work. You're debugging the
contenteditable injection timing on chatgpt.com's React SPA.

Why OpenRouter

CLAMP uses OpenRouter for distillation — not the chat platform's own API. This means:

• Model-agnostic — use Claude, GPT-4o, Gemini, or even free Llama models
• Your key, your cost — no CLAMP subscription, no data sent to us
• ~$0.001 per compression with Gemini Flash 2 (the default)

Roadmap

• Firefox support
• Chrome Web Store release
• clamp watch — token counter overlay that warns before you hit the wall
• Auto-detect session % and prompt compression proactively
• Offline mode — local distillation via Ollama

Contributing

The scraper selectors break when AI platforms redeploy. This is the #1 thing that needs community maintenance.

If CLAMP stops working on a platform:

1. Open DevTools → Inspector on the conversation
2. Find the selector for human/AI turns
3. PR to update src/scrapers/<platform>.js

---

Built because this problem was too annoying to ignore.