95% Token Savings
Verified & Reproducible
Current benchmark with 1,360 decisions proves Continuity is 20.1× more efficient than embedding decisions in CLAUDE.md. At this scale, CLAUDE.md needs 311,752 tokens — well beyond the context window limit and literally cannot run. See the math, run the tests yourself.
Token Comparison
Current benchmark: CLAUDE.md climbs to 311,752 tokens while Continuity stays at 15,476
The Scaling Problem
CLAUDE.md hits context limits. Continuity scales infinitely.
Token Growth: O(n) vs O(1)
CLAUDE.md grows linearly with decisions. Continuity stays constant.
CLAUDE.md hits the 200K context limit at about 864 decisions and becomes unusable.
At 1,360 decisions CLAUDE.md would require 311,752 tokens - about 155.9% of the context window. Continuity stays at 15,476 tokens for the current benchmark.
Calculate Your Savings
See exactly how much you'll save based on your project size
Calculate Your Savings
CLAUDE.md vs Continuity
They serve different purposes and work best together
CLAUDE.md
Static documentation file
Grows linearly with decisions
•Loads all decisions every time
•Manual editing required
•Breaks around 864 decisions
•No search capability
Best for: Project instructions, coding standards, static context (~5,201 base tokens)
Continuity
Intelligent memory system
Constant regardless of decisions
•Loads only what you need
•Automatic logging
•Unlimited decisions
•Semantic search
Best for: Decision memory, searchable history, dynamic queries (15,476 tokens at the current benchmark)
| Feature | CLAUDE.md | Continuity |
|---|---|---|
| Static project instructions | ||
| Store 100 decisions | ||
| Store 500 decisions | ||
| Store 864+ decisions | ||
| Search specific decisions | ||
| Semantic similarity search | ||
| Log decisions automatically | ||
| Session handoff | ||
| Decision relationships | ||
| Link decisions to code | ||
| Token efficient(O(n) vs O(1)) | ||
| Team collaboration |
Real-World Scenario: "What did we decide about authentication?"
With CLAUDE.md Only
- 1.Claude loads all 1,360 decisions (311,752 tokens — 155.9% of the 200K context window)
- 2.Claude scans through everything manually
- 3.You've already used 155.9% of context before even asking
- 4.Doesn't work - exceeds the context limit
With Continuity
- 1.Claude calls
search_decisions("authentication") - 2.Returns only the relevant decisions (~15,476 tokens at the benchmark)
- 3.You've used just 7.8% of context
- 4.Works with room for long conversations
The Verdict: They Work TOGETHER
Use CLAUDE.md for:
What to do (instructions)
Use Continuity for:
What we decided (memory)
CLAUDE.md and Continuity serve different purposes. They're not replacements - they're complements.
Scaling Projections
Token savings increase as your project grows
| Decisions | CLAUDE.md Tokens | Continuity Tokens | Savings | Status |
|---|---|---|---|---|
| 50 | 16,471 | 15,476 | 6.0% | |
| 100 | 27,742 | 15,476 | 44.2% | |
| 250 | 61,552 | 15,476 | 74.9% | |
| 500 | 117,904 | 15,476 | 86.9% | |
| 864LIMIT | 199,951 | 15,476 | 92.3% | |
| 1,000 | 230,606* | 15,476 | 93.3% | |
| 1,360ACTUAL | 311,752* | 15,476 | 95.0% | |
| 5,000 | 1,132,227* | 15,476 | 98.6% | |
| 10,000 | 2,259,252* | 15,476 | 99.3% |
Key Insight: Savings increase as your project grows.
The more decisions you log, the more efficient Continuity becomes. At 10,000 decisions, you save nearly 99.4% of tokens.
Mathematical Proof
Formal proof that Continuity provides asymptotically superior token efficiency
| Symbol | Definition | Measured Value |
|---|---|---|
| n | Total number of decisions | 1,360 |
| D | Total decision tokens | 306,551 |
| d | Average tokens per decision | 225.4 |
| b | Base CLAUDE.md instructions | 5,201 |
| q | Search queries per session | 3 |
| r | Results returned per query | 15 |
| C | Context window limit | 200,000 |
The Fundamental Equation
As decisions grow, Continuity's relative cost approaches zero.
Q.E.D.
Getting Started
What is Continuity?
Continuity is a RAG-based synthetic memory system for AI coding assistants. It captures architectural decisions, tracks documentation changes, and provides persistent context across sessions - solving the problem of AI assistants "forgetting" previous conversations.
Key Advantage: O(1) Token Efficiency
Unlike CLAUDE.md which loads ALL decisions every session (O(n)), Continuity uses semantic search to retrieve only relevant decisions - keeping token usage constant regardless of project size.
Installation
Option 1: VS Code Marketplace (Recommended)
- Open the Continuity listing on the VS Code Marketplace
- Click Install
- Reload VS Code when prompted
Option 2: Manual Installation
- Download the latest .vsix file
- Open VS Code
- Press
Cmd/Ctrl+Shift+P - Type "Extensions: Install from VSIX..."
- Select the downloaded file
- Reload VS Code when prompted
What Happens on First Launch
After installation, everything is automatic:
- Zero-click MCP setup — auto-detects your AI tools (Claude Code, Cursor, Cline, Roo Code, GitHub Copilot, Google Gemini, and any MCP-compatible client) and configures MCP connections
- Smart Node.js resolution — finds Node.js automatically for nvm, fnm, volta, and Homebrew users
- Initial Project Scanner — scans your codebase and creates draft decisions for tech stack, dependencies, database, testing, deployment, API style, and project structure
- Git seeding — pre-populates decisions from your recent commit history so the sidebar is never empty
- A
.continuity/folder is created with your decisions as plain JSON — commit it to git and share with your team
14-Day Free Trial
Your trial starts immediately with all Pro features and unlimited decisions. After the trial, you keep read-only access to existing decisions and can continue logging up to the free-tier limit. No credit card required.
Features
14-Day Free Trial
Every new install starts with a full 14-day trial — unlimited decisions, all Pro features. After the trial, the free tier includes 10 decisions (15 with email verification).
Unlimited decisions, semantic search, knowledge graph, Dream Engine
Current shipping profiles expose 9 core tools, 25 standard tools, or 36 full tools depending on the client
Auto-detects and configures all your AI tools
19 detection points auto-capture decisions
Pro Features
Upgrade to ProHybrid keyword + embedding-based search (O(1) efficiency)
1,500 node capacity — force-directed, radial, tree, and DAG layouts with inline editing
4-phase memory consolidation — deduplicates, resolves contradictions, scores staleness
Active, draft, outdated, deprecated, superseded — old decisions auto-outdated, never deleted
Link decisions to code locations, check refactoring safety against linked decisions
Pre-commit audit blocks if >5 unlogged, post-commit auto-logs stubs, decision debt tracker
No storage limits — in-memory cache with ~60% smaller trimmed MCP responses
CSV, JSON, Markdown formats
License Management
Activation
After purchasing, you'll receive a license key via email. Activate it in VS Code:
- Press
Cmd/Ctrl+Shift+P - Type "Continuity: Activate License"
- Enter your license key and email
License Types
Renews monthly, cancel anytime. 14-day free trial included.
Renews annually, save 18% vs monthly. 14-day free trial included.
Never expires, all future updates included
MCP Integration
What is MCP?
Continuity integrates with the Model Context Protocol (MCP) to provide AI assistants like Claude, Cursor, and Cline with persistent memory and decision logging capabilities.
Available MCP Tools
Continuity ships profile-based MCP tool sets for logging, searching, analyzing, and managing architectural decisions. Current profiles expose 9 core tools, 25 standard tools, or 36 full tools depending on the client and install mode.
@continuity log_decisionLog architectural decisions with question, answer, and optional tags
@continuity search_decisionsSearch logged decisions by keyword or semantic query - O(1) efficiency
@continuity get_quick_contextLoad project history, recent decisions, and session context at conversation start
@continuity session-handoffGet complete project context including recent decisions and documentation status
continuity://ai-usage-guideAI-specific usage guidelines and best practices
continuity://ai-instructionsPersistent AI instructions that survive across sessions
API Reference
License Validation
POST /api/licenses/validate{
"license_key": "12345678-90ab-cdef-1234-567890abcdef",
"device_id": "device-uuid",
"device_name": "MacBook Pro",
"platform": "darwin",
"vscode_version": "1.84.0",
"extension_version": "2.13.89-beta"
}Device Management
POST /api/devices/activateActivate a new device for a license
DELETE /api/devices/:device_idDeactivate a device
Troubleshooting
License activation fails
- Verify your license key is correct (check for typos)
- Ensure you're using the email from your purchase
- Check your internet connection
- Contact support if issue persists
MCP integration not working
- Verify Continuity is installed and activated
- Check that your AI assistant supports MCP
- Restart your AI assistant application
- Check MCP server logs in VS Code Output panel
Need more help? Email us at [email protected] with your license key and a description of the issue.
Frequently Asked Questions
Why is Continuity more efficient than CLAUDE.md?
CLAUDE.md loads ALL decisions into context every session (O(n) complexity - grows linearly). Continuity uses semantic search to retrieve only relevant decisions (O(1) complexity - constant). At 1,360 decisions, CLAUDE.md needs 311,752 tokens while Continuity uses just 15,476 tokens. That is a 95% token reduction with a 20.1× efficiency advantage, and CLAUDE.md is already beyond the 200K context window.
What happens when I have more than 864 decisions?
Around 864 decisions, CLAUDE.md reaches the 200K context window and stops fitting reliably. Continuity keeps working because it only retrieves the relevant decisions needed for the current session.
How much money will I save?
Based on the current benchmark at 600 sessions/month: CLAUDE.md costs $561.15/month while Continuity costs $27.86/month. That is $533.30/month or $6,399.56/year in benchmark-model savings.
Is my data secure?
Yes. All decisions are stored locally in your project's .continuity/ folder. License validation uses encrypted HTTPS connections, and we never store your decision data on our servers.
What AI assistants does Continuity work with?
Continuity works with any AI assistant that supports the Model Context Protocol (MCP): Claude Code, Cursor, Cline, Roo Code, GitHub Copilot, Google Gemini, and any other MCP-compatible client. Zero-click setup auto-detects your installed tools and configures MCP connections automatically.
Still have questions?
Our team is here to help. Reach out via email or check our activation guide for step-by-step instructions.