adash-cli 0.1.0

htop-style monitor for Claude Code and Codex CLI agent sessions

adash

AI agent monitor for Claude Code and Codex CLI — a local-first triage dashboard for session health, cost, and utility.

Inspired by and data format compatible with graykode/abtop (MIT). adash is a cleaner Python rewrite using Textual rather than a btop pastiche.

A sibling to mnemos, pulldown, cartograph, and synthesis.

---

What it shows

• Summary strip: active sessions, attention count, live burn/hour, today estimate, hottest context/quota, orphan count
• Ranked attention queue: title, project, model, ctx %, token total, estimated cost, burn/hour, age
• Inspector: overview, token/cost breakdown, recent trend, alerts, ports/subagents, and safe actions
• JSON snapshot output with raw sessions plus derived summary, alerts, and insights

Install

pipx install adash-cli
# or
pip install adash-cli

For local development and test runs:

pip install -e ".[dev]"
python -m pytest -q

If you want the latest repo state before a PyPI release lands:

pipx install git+https://github.com/anthony-maio/adash.git

Usage

adash              # TUI (default)
adash --once       # Snapshot table to stdout, exit
adash --json       # JSON snapshot to stdout, exit
adash --setup      # Print Claude StatusLine hook setup instructions
adash --version    # Print version

Keybindings

Key	Action
↑ / ↓ or j / k	Select session
enter	Copy session ID to clipboard
o	Open selected session cwd
t	Open selected transcript
x	Kill selected session (SIGTERM)
X	Kill all orphan port processes
f	Toggle all vs risky sessions
s	Cycle sort mode (severity, cost, recency)
r	Force refresh
q	Quit

Claude rate limit hook setup

adash reads ~/.claude/adash-rate-limits.json (also reads legacy ~/.claude/abtop-rate-limits.json for back-compat) for Claude quota data. This file is written by a shell hook you add to your Claude Code setup.

Run adash --setup to print the hook snippet. Add the function to your shell and configure it via Claude Code's hooks setting in ~/.claude/settings.json.

Codex rate limit data

Codex rate limit data is extracted directly from the session JSONL files that Codex writes during its token_count events. No hook needed — just run Codex normally and adash picks it up.

Rate limit cache is stored at ~/.cache/adash/codex-rate-limits.json.

Pricing overrides

adash includes built-in approximate pricing rules for common Claude and GPT-5 model families. To override them locally, create:

• $ADASH_PRICING_FILE, or
• ~/.config/adash/pricing.json

File shape:

{
  "models": [
    {
      "pattern": "claude-sonnet-4*",
      "agent_cli": "claude",
      "input_per_million": 3.0,
      "output_per_million": 15.0,
      "cache_read_per_million": 0.3,
      "cache_write_per_million": 3.75
    }
  ]
}

Estimated USD values are intentionally approximate and secondary to token/quota signals.

Data compatibility

adash reads the same session files as abtop:

• Claude sessions: ~/.claude/sessions/*.json + transcripts in ~/.claude/projects/
• Codex sessions: ~/.codex/sessions/YYYY/MM/DD/rollout-*.jsonl

The transcript encoding rule (Claude): /, _, . → -.

Attribution

Inspired by graykode/abtop (MIT License). Data format, JSONL parsing logic, and session discovery strategy are derived from abtop.

License

MIT — see LICENSE.

Development

• CI runs the test suite on Linux and Windows to keep the collector and path-handling logic portable.
• adash --json now returns structured JSON errors with a non-zero exit code when collection fails, which makes automation safer than relying on a traceback.