Continuum

Seamless AI session continuity — token analytics, context capsule handoffs, and prompt recommendations across all your AI coding tools.

npm install -g @hashbyharsh/continuum
continuum web

Opens a live dashboard at http://localhost:3737 showing your token usage, costs, and session health across Claude Code, Cursor, Cline, Copilot, and more.

Features

Token Analytics

Real-time cost and token tracking across every AI coding session — broken down by provider, model, project, and task category (coding, debugging, refactoring, planning, etc.).

Context Capsule

Automatically detects when a session is approaching its model's context window limit (warning at 70%, critical at 90%). Generate a .capsule.md handoff file with a single click — containing a structured summary, key decisions, open questions, and a ready-to-paste continuation prompt for a fresh session or a different model.

Prompt Recommendations

Analyzes your past prompts for token-wasting patterns and suggests drop-in templates to fix them. Detects: verbose prompts, repeated context, missing cache hints, vague task descriptions, and missing output format constraints.

Web Dashboard

A self-contained dark-mode dashboard with four tabs:

• Overview — KPIs, token charts by provider, cost by category, model breakdown table
• Sessions — all sessions sorted by recency with context utilization bars
• Context Capsule — alerts for sessions nearing their limit, one-click capsule generation
• Prompt Tips — ranked recommendations with copy-paste templates

Supported Providers

| Provider | Log Location | |---|---| | Claude Code | ~/.claude/projects/ | | Claude Desktop / Cowork | ~/Library/Application Support/Claude/local-agent-mode-sessions/ | | Cursor | ~/.cursor/ | | Cline | VS Code extension logs | | Roo Code | VS Code extension logs | | GitHub Copilot | VS Code extension logs | | Codex CLI | ~/.codex/ | | Gemini CLI | ~/.gemini/ | | OpenCode | ~/.opencode/ | | Kimi | ~/.kimi/ |

CLI Commands

# Launch web dashboard (default port 3737)
continuum web

# Print token stats in the terminal
continuum stats

# Generate a context capsule for a session
continuum capsule --session <sessionId>

# Show prompt recommendations
continuum recommend

# List all detected providers
continuum providers

# Filter by provider or date range
continuum stats --provider claude --days 7

How Context Capsule Works

When a session hits 70%+ of its model's context window, Continuum flags it. Clicking Generate Capsule produces two files in ~/.continuum/capsules/:

• <session>.capsule.md — human-readable summary with a continuation prompt
• <session>.capsule.json — structured data for tooling

Paste the continuation prompt into a new session to pick up exactly where you left off — even on a different model.

Requirements

• Node.js >= 18
• At least one supported AI coding tool installed locally

Install & Update

# Install
npm install -g @hashbyharsh/continuum

# Update to latest
npm install -g @hashbyharsh/continuum@latest

# Uninstall
npm uninstall -g @hashbyharsh/continuum

License

MIT © Harsh Singh