Argus – VSCode debugger for Claude Code sessions
Debugger for Claude Code session costs—step-by-step breakdown of tokens and tool calls.
claude code analysis terminal app
Sparklines, heatmaps, and chat replay for Claude Code—local, no data exfil.
Claude Code users who need local session analysis and usage insights
VisiData · Atuin (shell history) · Local LLM usage dashboards
How to install: brew install base-14/tap/cicada or go install github.com/base-14/cicada@latest then > cicada
what it shows -
├── Analysis │ ├── Usage heatmap │ ├── Sessions per day (sparkline) │ ├── Messages & tools per session (bar charts) │ ├── Streaks (current, longest, weekly) │ ├── Personal bests (longest session, most messages, most tools) │ └── Trends (sessions this week vs last, avg duration) │ ├── Projects │ ├── All projects with session counts and last active │ └── Project detail (Enter to drill in) │ ├── Overview — total sessions, messages, duration │ ├── Sessions — per-project session list │ ├── Tools — tool usage breakdown for this project │ ├── Activity — project-level heatmap │ └── Skills — which skills were invoked │ ├── Sessions │ ├── All sessions with project, duration, messages, cost │ └── Session detail (Enter to drill in) │ ├── Chat — full conversation: user prompts & assistant responses │ ├── Overview — duration, message count, model, cost │ ├── Timeline — chronological tool calls and messages │ ├── Files — files read, written, and edited │ ├── Agents — subagent spawns and results │ └── Tools — per-session tool call breakdown │ ├── Agents │ └── Subagent usage across all sessions │ └── Tools ├── Built-in tools ranked by call count └── MCP server tools with server grouping
Debugger for Claude Code session costs—step-by-step breakdown of tokens and tool calls.
Single gist script parsing local JSONL when no official Claude analytics exist.
Cuts 1.5M tokens from bloated Claude sessions with compaction archaeology and cost-per-decision tracking.
Finally see which subagent spawned what—Claude Code and Codex in one view.
Debugger for Claude Code sessions—niche but solves real frustration with cost visibility.
First analytics layer for Claude Code revealing 26% session abandonment rate.