Skip to content

OpenTime

Temporal awareness and time-effort estimation for AI agents.

The Problem

AI agents have no concept of time. Their time intuition comes from LLM training data calibrated to human speed, not agent speed. This causes real problems:

  • Bad timeouts — An agent sets a 60-minute timeout on a download that takes 65 minutes. It times out at 98%, losing all progress.
  • Wrong decisions — An agent chooses a 23-hour approach over a 9-hour one because it estimates coding at human speed. But the coding takes minutes, not hours.
  • No self-awareness — Agents don't know how long they take to complete tasks, how long they've been running, or when the last user interaction was.

The Solution

OpenTime gives any AI agent the ability to:

  • Track time — wall clock, elapsed time, stopwatches
  • Record events — task start/end with automatic correlation IDs
  • Learn duration estimates — per-agent statistics (mean, median, p95)
  • Get timeout recommendations — data-driven timeouts based on history
  • Compare approaches — choose the fastest path using actual speed data
  • Passive tracking — automatically record tool usage via hooks

How It Works

Your AI Agent
    │
    ├── MCP Server (Claude, Cursor, Windsurf, Cline, Zed, etc.)
    ├── REST API (ChatGPT, Gemini, LangChain, any HTTP client)
    ├── LangChain Tools (native integration)
    ├── OpenAI/Gemini Functions (zero-dependency schemas)
    └── Passive Hooks (Claude Code, Cursor, Cline, Copilot, Windsurf, Amazon Q)
    │
    ▼
OpenTime (per-agent SQLite database)
    │
    ▼
Duration Statistics → Timeout Recommendations → Decision Support

Quick Start

pip install opentime

Then choose your integration:

  • MCP Server — For Claude Code, Cursor, Windsurf, and other MCP clients
  • REST API — For any agent that can make HTTP calls
  • Docker — One command, no Python needed
  • Passive Hooks — Fully automatic, no agent action needed

See the full Getting Started guide.