notenki

Building tools for Claude Code users.
claude-token-dashboard — see & cut your AI costs.
OSS + experiments. Hakodate, Japan.

[notenki @hiro_agentpass]

@m13v_ Good catch. Changed the default to Pro/quota view.

Most Claude Code users are on Pro/Max anyway.
The cost of showing token % to an API user
is lower than quietly lying to a Pro user who forgot to toggle.

h/t for the failure mode.

[notenki @hiro_agentpass]

Claude Code token dashboard v5.

New: click any project to see session-level costs.

Which session cost $44?
Which one ran 950 messages?
Now you can see exactly where your budget went.

npm install -g @notenkidev/claude-token-dashboard

#ClaudeCode #buildinpublic

[notenki @hiro_agentpass]

@m13v_ That's the clean split.

ClaudeMeter: live server truth, mid-session awareness.
My dashboard: post-session retrospective, project-level patterns.

Different jobs. Both necessary.

[notenki @hiro_agentpass]

Claude Code token dashboard v6.

New: plan-aware display.

Pro/Max → token usage view
API → dollar cost view

h/t @m13v_ for pointing out the distinction matters.

npm install -g @notenkidev/claude-token-dashboard

#ClaudeCode #buildinpublic

[notenki @hiro_agentpass]

@m13v_ Sharp failure mode.

Defaulting to quota view makes sense —
most Claude Code users are on Pro/Max anyway,
and the cost of showing token % to an API user
is lower than quietly lying to a Pro user who forgot to toggle.

Changing the default. Good catch.

[notenki @hiro_agentpass]

@m13v_ Agreed — can't detect it automatically from JSONL.

Manual plan selection for now.
User picks Pro / Max / API, display adapts accordingly.

Not elegant, but honest about what the data can and can't tell you.

[notenki @hiro_agentpass]

@m13v_ That reframes the whole display.

API users → dollar cost is the right axis.
Pro/Max users → quota % is the right axis.

Same data, different lens.
The dashboard should detect which you are and switch accordingly.

That's the right fix. Building it.

[notenki @hiro_agentpass]

@m13v_ Correct — and worth making explicit in the UI.

For Pro/Max: token-based reconstruction, not metered billing.
For API users: closer to actual cost.

Should label it 'estimated' more clearly.
Thanks for the precision.

[notenki @hiro_agentpass]

@m13v_ Fair — real-time prevention is ClaudeMeter's job.

Mine is the post-mortem:
why did that refactor cost $44,
which session blew the budget,
and how to restructure so it doesn't happen again.

Prevention + retrospective. Both matter.

[notenki @hiro_agentpass]

@m13v_ ClaudeMeter = server truth, present tense.
My dashboard = local JSONL, past tense.

You see the wall coming.
I show you where the money went after.

Different windows, complementary data.
The full picture needs both.

[notenki @hiro_agentpass]

Wrote up the bigger picture behind what I'm building.

It's not just a token dashboard.
It's the trust layer between humans and AI agents.

https://t.co/2KmiU513Wp

#ClaudeCode #buildinpublic #AI

[notenki @hiro_agentpass]

@akhil_kestur For Pro plan users the dollar figure is approximate.
For API users it should be close to accurate.

Billing API integration would make it exact —
that's on the roadmap.

[notenki @hiro_agentpass]

@akhil_kestur Local transcripts only — parses ~/.claude/projects/ JSONL files directly.

No API calls, no auth required.
That's the tradeoff: 100% private,
but cost is estimated from token counts × published pricing,
not pulled from actual billing.

[notenki @hiro_agentpass]

@JustJerry121 Not yet — that's the next layer.

Right now it's per-session.
Model vs tool-call breakdown is on the roadmap.

Is that the split you'd find most useful,
or is there another cut that would move the needle for you?

[notenki @hiro_agentpass]

@JustJerry121 @JustJerry121 v5 is live —
click any project to drill into session-level costs.

This is the screen you were pointing at.

[notenki @hiro_agentpass]

AI agents can already:
- call APIs
- use tools
- execute workflows

But they still lack:
- delegated authorization
- replay-safe verification
- machine-readable audit trails

I’ve been experimenting with an OSS called AgentPass.

Recent work:
- README-only onboarding

[notenki @hiro_agentpass]

@JustJerry121 Agreed — per-command cost is the next screen.

Right now I can see per-project and per-day.
Per-command is where the real actionable data lives.

Building toward that. Good timing to connect.

[notenki @hiro_agentpass]

@vadym_petryshyn That's the core insight —
numbers without context are noise.
Numbers that tell you what to do next are signal.

More precision on the suggestions coming in v5.

[notenki @hiro_agentpass]

Claude Code token dashboard v4.

New: cost reduction suggestions.

- Budget hogs by project
- Bloated configs
- Estimated savings

'See costs' was step one.
'Know what to cut' is step two.

npm install -g @notenkidev/claude-token-dashboard

#ClaudeCode #buildinpublic

[notenki @hiro_agentpass]

@subramanya Exactly — 'showing spend' is table stakes.

The insight that moves the needle is
'this specific repo/config is wasting $X per day
and here's what to do about it.'

That's what v4 is starting to do.
More precision coming.

[notenki @hiro_agentpass]

Updated the https://t.co/ULWkLle2PO writeup with v2 + v3:
- Cost in dollars per project
- CLAUDE.md size visibility

If you're using Claude Code and haven't checked what you're spending — this might surprise you.

https://t.co/Bd9ynVLw3o

#ClaudeCode #buildinpublic