Local token reports

llm-tokei

Turn local agent sessions into clear token, cache, and cost reports.

$ cargo install llm-tokei

Why

One report for many agents.

Usage is scattered across IDE chats, terminal sessions, databases, model names, and pricing tables. llm-tokei makes it one local report.

Default view source x model, sorted by total

Time windows 24h, 7d, month, or exact dates

Scriptable output tables for people, JSON for scripts

Session dumps Codex and Copilot JSONL

Agent Default location Artifact

Codex CLI $CODEX_HOME/sessions session JSONL

OpenCode $OPENCODE_DATA_DIR/opencode.db read-only SQLite

Claude Code $CLAUDE_HOME/projects project logs

Copilot Chat editor workspaceStorage chat sessions

Copilot CLI ~/.copilot/session-state event logs

Counts

prompt and cache tokens

assistant tokens

reasoning tokens when available

mixed, official, and provider splits

distinct local histories

Month by provider

llm-tokei --cost-per provider --cost official --month -h

Project last day

llm-tokei --24h --group-by project,source,model

Expensive sessions

llm-tokei --7d --group-by session,source,model --sort cost --limit 10

Automation JSON

llm-tokei --7d --format json --group-by source,model,project

Local

llm-tokei reads local artifacts and opens OpenCode data read-only. No provider history API required.