@latten/glow

Local, zero-egress visibility into what your coding agents cost — and a calm nudge to take a break. See your whole AI-life balance across every agent with latten balance, and get a quiet per-turn line in Claude Code via the hook. Nothing leaves your machine — no account, no network call.

Your balance across every agent

latten balance     # ltn balance also works

One view of your coding-agent usage today, unified across the agents whose local sessions Latten can read — Claude Code and Codex today (Gemini is detected, but its local logs don't expose token counts yet) — with a nudge to take a break and stretch your legs when you've done enough:

  ▌ today across your agents

    claude code   251.7M tok
    codex           3.2M tok
    ──────────────────────────────
    today         254.9M tok

    Big day. You earned the quiet that comes after.

It reads each agent's local session files — only the ones you already use — counts tokens, and gets out of the way. Types and counts, never your prompts, code, or data. You choose what, if anything, to share.

From your balance to a team graph

Glow is free and local. When Latten sees that the current repo has more than one contributor, it may show a quiet team nudge: your local balance is useful, but a shared Latten graph shows the whole team's AI cost and reach in one place. That team graph is powered by @latten/installer and @latten/sdk, not by uploading your local Glow history.

To create that graph, connect @latten/installer as an MCP server and let your agent call provision, instrument the repo, verify real crossings, and hand you the receipt plus claim link. The receipt is useful on its own. Claiming keeps the history and opens the paid path for recommendations, fix PRs, and verification.

Install the per-turn line (Claude Code)

npx @latten/glow install

This writes a Stop hook entry into ~/.claude/settings.json. Claude Code calls it automatically at the end of every turn.

What the per-turn line means

latten · this turn ~$0.03 · 12,400 tok.

• ~$0.03 — estimated USD cost for the last turn based on the model and token counts in the transcript. null if the model is unrecognised (never guesses).
• 12,400 tok — total tokens (input + output + cache read + cache creation).

If a waste pattern is detected, you'll see a description instead:

latten · 4 turns looked redundant (~$0.12). run `latten why` for detail.

Uninstall

npx @latten/glow uninstall

Removes the hook entry from ~/.claude/settings.json. Your Claude Code session is unchanged otherwise.

You moved fast with AI. Latten helps you stay in control of it. See what your AI costs — and where your data goes: Latten turns the AI calls across your company into one live graph — cost, reach, and PII exposure, attributed and observed from real traffic. One prompt to your coding agent, a PR you review, live in minutes. latten.io