The AI coding meter for every agent you run.
Ottto reconciles subscriptions, API usage, credits, quota windows, and cloud bills into one number you can defend. No proxy in the request path, no added latency, no mystery totals. Add another AI tool without adding another thing to watch.
Claude Code 
Codex 
Cursor 
Pi 
Gemini 
Ghostty 
iTerm 
cmux 
openclaw 
Hermes + more We don't want to spend less on AI. We want to scale up without flying blind.
We found ourselves using 3x the AI of last month, across Codex, Claude Code, Pi, Bedrock, Vertex, subscriptions and APIs, on multiple Macs. The bill stopped being background noise. So we keep checking the same things, several times a day:
Where are we losing efficiency to huge context, bad caching, noisy MCP tools, retries, or sessions we reopen after days idle?
Advisor finds itHow much more are we paying now that agents run loops and spawn other agents, instead of us driving every step?
agent vs humanWhich agents are expensive but not productive?
cost by toolHow much are we paying in APIs right now, and would a fourth $200 subscription be smarter?
subs vs PAYGWhat is our total AI coding cost?
one numberThe bill stopped being background noise.
AI is the fastest-growing line in most engineering budgets - and a single agent task now spends like a small invoice. This isn't a rounding error anymore.
Not per device. Not per account. One reconciled truth layer.
Every dollar, reconciled. Ottto pulls usage from every agent, every account, on every machine - plus the cloud and your credits - into one trusted view. By tool, model, account, session, repo, and billing path. Subscriptions, API usage and credits, side by side.
- ✓All your accounts. A personal and a business Codex, two Claude plans + API, all deduped into one number.
- ✓All your agents and machines. Claude Code, Codex, Cursor, Pi - and the cloud - together, not one at a time.
- ✓Subscriptions, API and credits. Monthly plans, metered usage, and promo credits in the same picture.
- ✓Every number labeled. Exact where we know it, estimated where we don't.
- ✓One glance, the bottom lines. Every subscription and API at a glance - on your Mac or your phone. No tab-switching, no mental math, and a heads-up when you're pacing over.
- ✓Your data, your agents. The unified data layer is exposed over CLI, MCP, plugins and skills - so your own agent can pull last week's real usage, line it up against your merged PRs, and check whether the spend is turning into shipped code.
Claude · 2 plans + API$1,340exactCodex · Pro + Business$540exactMost of your spend is productive. Some of it buys nothing.
You're not overpaying for speed, that's the point. But some spend buys nothing back: re-paid context from broken caching, a plan on the wrong tier, a 1M window where 200K would do, or output tokens - billed several times higher than input - spent on verbose intermediate chatter no one reads. Ottto separates the spend that makes you faster from the spend that doesn't, and shows you exactly how much - so you can act, like keeping models terse in intermediate steps and verbose only in the final answer.
A short queue of fixes, not another dashboard.
Advisor watches quota windows, model/cache behavior, long-context sessions, agent-spawned work, and plan fit every day. It shows the dollars, the evidence, and the caveats before anything changes.
Set a budget per tool or per person and Ottto pings you on Slack or desktop while you can still act - before the invoice, not after.
Codex API spend is rising while a Claude Max window has room.
detector Route overflow away from paid API usage before the next reset.
detector Reprice the same traffic with the cache hit-rate drop called out.
detector Spot sessions where 1M context is being used where a smaller window would do.
detector Separate human-driven work from agents spawning other agents.
autopilot Demote, promote, or switch only when the usage evidence supports it.
Not a scary autopilot that moves money on its own. It finds, explains, prices, asks, executes only approved actions, and tracks rollback state.
Know before you hit the wall.
Every plan and account you pay for, how much you actually use, and when the next limit resets - so you're never surprised. Which plan fits, what you'd save on pay-as-you-go, and when to cancel before a renewal you won't use.
You won't refresh four dashboards at midnight, mid-task. Ottto watches every window - across machines, accounts and subscriptions - and warns you before you hit the wall.
- ✓Quota resets. "Claude Max - 78% of your 5-hour window. Resets 3:40pm."
- ✓Alerts before the wall. Set a threshold and get a heads-up before a window resets or a limit hits - not after you're blocked.
- ✓Renewal warnings. "Codex Business renews Tuesday - idle 3 weeks. Cancel?"
- ✓Savings vs pay-as-you-go. The cross-account math vendor dashboards never show.
The moment a provider changes, you know what it costs you.
Anthropic ships a new default. OpenAI changes pricing. A model gets deprecated, a quota window shifts. Ottto watches provider changes and reconciles each one against your actual usage so a price hike or default swap shows up as "here's what it does to your bill."
- ✓Real-time alerts. Pricing, defaults, deprecations, quota and outage changes.
- ✓Tied to your spend. Not "something changed" - "this changed your bill by $X."
The one place that sees all of it.
A vendor's dashboard charts one account - it won't reconcile your other plans, and it won't tell you what to change. Free meters watch one machine. Gateways sit in your request path and see only proxied API traffic. Ottto ties every device, agent and account together - and tells you what to do about it.
| Free CLIs & menu bars | Gateways | Vendor dashboards | ottto | |
|---|---|---|---|---|
| All your devices | one machine | proxied only | n/a | yes |
| All accounts & agents | one or a few | API only | one tool | yes |
| Subs + API + credits | quota only | usage only | its own | all |
| Reconciles every source | no | no | no | yes |
| Real-time provider-change alerts | no | no | no | yes |
| Recommends & helps act | no | limits | no | yes |
| In your request path | never | yes | n/a | never |
The things you're wondering.
QCan't I just check my usage myself?
You could - across every tool, account and machine. But you won't, not mid-task when you're deep in a problem, which is exactly when a limit hits or a bill creeps up. Ottto watches the windows for you and flags it before the wall.
QDoes Ottto slow down my coding tools?
No. Ottto is out of your request path - it reads telemetry your tools already emit. Zero added latency, and nothing breaks if Ottto is down.
QCan you see my code or prompts?
Ottto is built around usage, cost, and limit signals - not your source code. You choose what each source shares, and local data stays labeled and under your control.
QMultiple accounts and plans?
Yes - that's the point. A Pro and a Business Codex, two Claude plans, an API key and some credits all reconcile into one deduped picture across every Mac.
QDo I need an API key?
No. Ottto works with subscriptions too - Pro, Max, and the rest. Subscriptions, API and credits, all first-class.
QIs my subscription safe?
Yes. Ottto observes; it never proxies your requests, never holds your login, never limits which tools you use, and never touches provider terms - it just collects in parallel. So it isn't affected by the gateway restrictions vendors introduced in 2026.
QWhat is free and open source?
You can start free, no credit card. The runtime and provider feed are open source: ottto and ai-provider-watch.
"Not because I want to slow down. Because I want to scale up without flying blind."
The bill stopped being background noise, so I'm building the tool I kept wishing I had: open-source building blocks for the basics, and one product for the actionable bottom line. Usage and cost intelligence for AI coding agents, so you can push harder, not slower.
See your real numbers in minutes.
No sales call. No proxy. No slowdown. Install the app, connect your accounts, and watch every agent, machine and credit reconcile into one picture.