AgentCost

Tracking AI agent costs means switching between dashboards or checking each tool separately. I hit my usage cap with no clue which project burned through it — AgentCost puts it all in one place, right in the terminal where I work.

Unified cost visibility across Claude Code and Codex without leaving my workflow. CLI-first so agents can query costs programmatically via plain text or --json output, and a Bubble Tea TUI for interactive exploration.

Incrementally parses local JSONL logs, applies cache-aware pricing (separate rates for input, output, cache read, and cache creation), and stores normalized usage in SQLite. Daily/weekly/monthly breakdowns, per-model comparisons, per-project attribution derived from log directory paths, and live usage-limit gauges for both Claude and Codex via their OAuth APIs.

Built in Go with the Charm ecosystem.