Lancetta

Product Introduction

1. Definition: Lancetta is a native macOS menu-bar application (Menu Bar Extra) built with SwiftUI. It functions as an intelligent quota management and workflow optimization assistant specifically for AI-powered coding agents, currently supporting Anthropic's Claude Code and OpenAI's Codex.
2. Core Value Proposition: Lancetta exists to solve the cognitive overhead and quota waste associated with managing AI coding assistants. It transforms passive quota monitoring into actionable, data-driven guidance. Its core innovation is providing "when to push on, slow down, or switch agents" suggestions derived from historical quota data and system analysis, all without spending a single token on queries, ensuring zero-cost intelligence.

Main Features

1. Intelligent Quota Guidance Engine: This is Lancetta's core functionality. It continuously polls and stores quota usage data (percentage and reset timers) for both Claude Code and Codex. Using this time-series data, it calculates usage rates and projects future states. It then generates context-aware suggestions like "spend unused quota before a window resets" or "slow down usage pace to avoid hitting a limit." Each suggestion displays the underlying data (e.g., "5 hours: 82% used, 21% an hour over the last 1h00m, from 14 readings") for transparency.

   • How it works: The app reads local agent state files and account APIs directly. For Codex, it uses command-line queries that return account data without consuming tokens. For Claude Code, it leverages the existing sign-in session file on the macOS keychain, mimicking Claude Code's own /usage command, avoiding model calls.

2. Agent Setup & Maintenance Scanner: Lancetta performs a static analysis of the instruction files (e.g., CLAUDE.md, AGENTS.md, rules, skills) that AI agents load at the start of every session, which incur token costs. It scans recent repositories and parent folders, measures the token load against Claude Code's internal warning thresholds, and identifies 22 specific checks for optimization, such as overly large context files or redundant instructions.

   • How it works: The feature parses file structures and contents in designated workspace directories. It applies heuristics based on known agent performance impacts and provides actionable fixes—some automated, others with pre-written prompts to open directly in the agent for manual refinement.

3. Unified Session Wall & Orphaned Process Reaper: This feature addresses multi-session workflow management and system resource cleanup. It aggregates all active Claude Code and Codex sessions into a single visual "wall," prioritizing sessions waiting for user input. Separately, it identifies and lists background process trees left behind by terminated agent sessions, which macOS and the agents themselves do not automatically clean up, allowing users to reclaim held memory (often gigabytes).

   • How it works: The session wall reads the session lists maintained by each agent's native process. The process reaper uses system-level process inspection (like ps and lsof) to find orphaned subprocesses related to the agents' workspaces, presenting them for selective termination.

Problems Solved

1. Pain Point: Inefficient AI Agent Quota Utilization. Users waste quota by not spending before a reset or inadvertently hitting limits mid-task, causing workflow interruption. Standard quota displays are passive percentages without predictive guidance or historical context.
2. Target Audience: Professional developers and technical teams who rely heavily on Claude Code and/or Codex for daily coding, code review, and debugging. Power users who run multiple, concurrent agent sessions and need to manage costs and context switching effectively.
3. Use Cases:
   • A developer starting a long refactoring task can check Lancetta to see if a 5-hour quota window is about to reset with 80% unused, advising them to "push on" and start immediately.
   • A team hitting their weekly Claude Code limit on a Tuesday uses Lancetta's trend analysis to see the impending limit days in advance and "slow down" by switching to Codex or lowering effort settings.
   • A user plagued by slow system performance discovers and uses Lancetta's process reaper to clean up 2.5 GB of memory held by dozens of orphaned agent processes.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros:
     • Zero-Cost Intelligence: The foundational advantage. Quota analysis costs no tokens, making frequent checks free.
     • Predictive, Actionable Insights: Moves beyond monitoring to prescriptive suggestions based on time-series data.
     • Deep macOS Integration: A true native menu-bar app (SwiftUI) with Notch support, no Electron overhead, and deep system integration for process/session management.
     • Transparent Logic: Every suggestion shows the data and calculations behind it, building user trust.
     • Comprehensive Workflow Aid: Combines quota management, session management, and system maintenance in one tool.
   • Cons:
     • Limited Agent Support: Currently only supports Claude Code and Codex for quota guidance. Cannot be extended by users to new agents.
     • macOS Exclusivity: Only available for macOS 15+, excluding Windows and Linux developers.
     • Complexity for Casual Users: The depth of features (process reaping, maintenance scans) may be overwhelming for users who only want basic quota visibility.

2. Key Alternatives & Differentiation:

   • Built-in Agent Quota Displays (Claude Code /usage, Codex UI): These provide only real-time, point-in-time data without history, projection, or advice. Lancetta differentiates by adding the critical dimension of time and predictive analytics.
   • General API Usage Dashboards (OpenAI Platform, Anthropic Console): These are web-based, aggregate all usage (not per-window), and lack integration with the local coding workflow and session state. Lancetta is context-aware, local, and focused on the developer's immediate decision-making.
   • Hypothetical "Generic AI Monitor" Chrome Extensions: Such tools might infer limits from transcript history, which is an estimate, not the official quota. Lancetta reads the official source directly and maintains a precise, persistent historical record. Its native app status also allows for system-level features (process management, menu-bar integration) impossible for a browser extension.

Frequently Asked Questions (FAQ)

1. Does Lancetta use my API tokens or quota to check my usage? No. This is its core technical differentiator. Lancetta reads quota data directly from local agent state and through official account query commands that do not invoke the AI models. For Codex, the lifetime token counter verifiably does not increment. For Claude, it uses the same authenticated request as Claude Code's own usage command, bypassing the model.

2. Can I use Lancetta with Cursor or GitHub Copilot? Lancetta's intelligent quota guidance and suggestions currently only work with Claude Code and OpenAI Codex. However, its Maintenance scanner feature does analyze instruction files for Cursor, GitHub Copilot, and Gemini to help optimize their context loading costs. Full quota support for additional agents depends on future app updates.

3. How does Lancetta's "when to slow down" suggestion work for weekly limits? Lancetta tracks not just the rolling 5-hour window but also the separate weekly limit that Claude Code enforces. By analyzing your daily usage rate against the remaining weekly quota and time, it can project—sometimes days in advance—if you are on pace to exhaust the weekly limit prematurely. It then suggests throttling measures (lower effort, model switching) to avoid running out mid-week.

4. Is Lancetta safe? Does it send my data anywhere? According to its documentation, Lancetta is a local-first application. All quota reading, analysis, and session tracking occurs on your Mac. The app is signed and notarized by Apple for security. You can review its privacy policy for specific data handling details, but the core functionality requires no external data transmission.

5. My Codex installation isn't detected. How do I fix this? Go to Settings › Sources within Lancetta. This pane lists all the standard installation paths Lancetta checked and what it found. If your codex binary is in a non-standard location (e.g., installed via a custom package manager), you can manually browse to and select the correct binary path in this settings menu.