Collection · 25 entries
Observability for AI coding agents
Tools that show you what your coding agents are actually doing: token spend, session state, tool calls, and parallel execution.
Coding agents are still mostly black boxes if you don't instrument them. The tools below give you visibility at different scales: per-session token cost (codeburn), live runtime state (abtop, lazyagent), session-graph visualization (agent-flow), and side-by-side comparison across engines (parallel-code). Pick by the question you're trying to answer - 'where is the spend going' is different from 'what is this agent doing right now'.
unclog - find and fix Claude Code context bloat
CLI that scans ~/.claude for MCPs, skills, hooks, and CLAUDE.md files, shows how many tokens each consumes before you type, and lets you remove items with reversible snapshots.
Recall - TUI search across agent session history
Local-first Rust TUI that searches Claude Code, Codex, and OpenCode session history with hybrid full-text plus semantic retrieval. Built on ratatui.
jeeves - TUI for browsing AI agent sessions
Terminal UI to search, preview, read, and resume Claude Code and Codex sessions in a unified view. More framework integrations planned.
claude-code-organizer - dashboard for CC configs
npx dashboard to manage Claude Code memories, configs, and MCP servers. Includes a tool-poisoning scanner, context token budget tracker, duplicate cleanup, and scope management.
coding_agent_session_search - 11-provider session search
Rust TUI and CLI that indexes and searches local coding-agent session history across Codex, Claude Code, Gemini, Cursor, Aider and seven other providers.
agent-sessions - native macOS session cockpit
Swift macOS app that browses, archives, and resumes sessions across Codex CLI, Claude Code, OpenCode, Gemini CLI, Factory Droid, GitHub Copilot CLI, and OpenClaw with rate-limit tracking.
opensessions - tmux sidebar for coding agents
tmux sidebar that tracks Amp, Claude Code, Codex, and OpenCode sessions with per-thread markers, a local HTTP API, and live session state.
horizon - GPU terminal canvas for agent sessions
Rust terminal app that puts every Claude/Codex/Gemini/Kilocode session on a GPU-accelerated infinite canvas you can pan and zoom across.
token-ninja - local shell command router for AI tools
MCP-fronted router that runs deterministic shell commands locally with ~19us latency, bypassing the LLM entirely. Cuts token spend on routine commands inside Claude Code, Codex, Cursor, and Copilot.
claude_agent_teams_ui - kanban for AI engineering teams
Kanban-style UI where agents take tickets, message each other, and review each other's work. Supports Codex, Claude, OpenCode, and 75+ LLM providers.
MioIsland - macOS Dynamic Island for Claude Code
Swift menu-bar app that surfaces Claude Code sessions in the macOS notch: monitor progress, approve permission prompts, and jump back into a session without leaving the current window.
onWatch - real-time agent API quota tracker
Lightweight Go daemon that tracks AI API quotas across Anthropic, Codex, Copilot, Antigravity, Synthetic, and Z.ai with a Material Design 3 dashboard. Local SQLite, zero telemetry.
claude-code-statusline - configurable Claude Code statusline
Statusline replacement for Claude Code with real-time cost tracking, MCP server status, and themeable layouts. Bash-based and cross-platform.
snip - LLM token-saving proxy in Go
CLI proxy for Claude Code, Cursor, Copilot, and Gemini that strips noise from LLM context with declarative YAML filters. Reports 60-90% token savings on typical agent traffic.
claudetop - htop for Claude Code sessions
Real-time terminal monitor for Claude Code: cost, cache efficiency, model comparison, and alert thresholds. Targeted at users running long agent sessions who need spend visibility.
TokenTracker - cross-tool token usage dashboard
Local-first tracker for Claude Code, Codex, Cursor, Gemini, Kiro, OpenCode, and OpenClaw token usage with a web dashboard, native macOS menu bar app, and four desktop widgets. Zero-config setup.
claude-code-usage-bar - statusline budget readout
Real-time Claude Code statusline showing token usage, remaining budget, burn rate, and depletion time. Drop-in shell prompt addition.
claude-inspector - visualize Claude Code prompt mechanics
Electron desktop app that visualizes how Claude Code constructs prompts under the hood. Useful for debugging context, tool calls, and prompt assembly when an agent goes off the rails.
claude-code-history-viewer - browse session history
Desktop app to browse and analyze Claude Code conversation history. Also reads opencode and Codex transcripts so you can compare runs across harnesses.
ContextAnalyzerTerminal - per-tool-call context analyzer
Python tool that breaks down Claude Code context window usage per tool call. Surfaces which Read or Grep calls are eating the most tokens.
prism - session intelligence for Claude Code
Python TUI that finds where extra tokens are burned in Claude Code sessions, why sessions fail, and what to fix. Built on Textual, focused on debugging your own usage.
Codeburn - token cost TUI for Claude Code & Codex
Interactive terminal dashboard that breaks down where your AI coding tokens actually go. Surfaces the chat-vs-tool-use split most users get wrong.
abtop - htop for AI coding agents
Rust TUI that monitors Claude Code and Codex sessions: token spend, context window, rate limits, and ports in real time. Like btop but for agent runtimes.
lazyagent - terminal observability for coding agents
Watch your Claude Code, Codex, and OpenCode sessions from a tiny Go TUI. Shows what each agent is doing, no SaaS dashboard needed.
agent-flow - real-time Claude Code orchestration view
Watch your Claude Code agents think, branch, and coordinate as they run. Live visualization of subagent trees and tool calls in the browser.
Related collections
Claude Code tools, plugins, and integrations
The best tools, MCP servers, and harnesses for getting more out of Claude Code - orchestration, observability, telemetry, and remote control.
Terminal UIs (TUIs) for daily workflows
Polished terminal UIs for git, tasks, observability, and agent control - the tools that make the terminal feel like a real surface again.
Multi-agent frameworks and orchestration
Frameworks, harnesses, and DSLs for coordinating multiple AI agents across handoffs, parallel waves, and tool use.