22. Billing and spend

The Billing page is one of the most-loved pages in Daalu. It puts cloud spend, local GPU usage, commercial LLM spend, and your Daalu subscription on a single screen, so the answer to “what does running this stuff cost?” is one click away.

URL: https://ops.daalu.io/billing. Admin-only.

Layout

Top of page: a summary band with three numbers.

• This month so far — total spend across all connected clouds.
• Last month — total at the same date in the prior month (for comparison).
• Projected month-end — straight-line extrapolation; not to be relied on for tight budgets, but useful for trend spotting.

Below, four sections:

1. Local GPU + LLM router
2. Cloud spend
3. Daalu subscription
4. Spend breakdown and forecast

We’ll cover each.

Local GPU + LLM router

A panel showing the status and value of local inference.

Status pill

• Online — local GPU has responded to a router probe in the last 5 minutes.
• Deploying… — Daalu is rolling the serving stack out onto your cluster; the collapsible Deploy progress panel below the pill shows the rollout live (pod state, blocking reasons, cluster events, and the serving logs once they exist).
• Offline — last response is older. Probably means the serving pod is unhealthy or the federation tunnel is down — the same Deploy progress panel is offered here too, because its events and logs are how you find out why.
• Not configured — no local GPU set up. The router is using its fallback tiers for everything. This is where the Connect your GPU button lives — chapter 35 walks through it.

The pill and the AI Factory page share one health probe, so they always agree.

SKU picker

Four cards showing the commercial equivalent of each tier of LLM call:

• Classifier (small) — short prompts, low context.
• Chat (default) — typical assistant turns.
• Chat (large context) — long-context summarization.
• Embedding — vector embeddings.

Each card has the commercial $/1M tokens price and the estimated “what you saved this month” figure.

Inference trend

A stacked bar chart of inference calls per day for the last 7 days, broken down by destination (local vs cloud) and tier (classifier / chat / embedding).

The first time you see this on a tenant with local inference turned on, the proportion of green (local) to gray (cloud) is usually a happy moment.

Spend

Total LLM spend this month, with the comparison “would have been $X without local inference.”

Cloud spend

A row per connected cloud account showing:

• Cloud name.
• This month’s spend so far.
• Last month’s spend at the same date.
• Trend arrow.

Click a row to drill into:

• Spend over time (last 90 days).
• Top services by spend.
• Top regions by spend.
• Top tags by spend (if you tag religiously).
• Largest individual resources (top-10 EC2 instances, top-10 S3 buckets by storage, etc.).

The data comes from the cloud’s billing API. Latency from “resource exists” to “shows up here” is typically 24 hours; spend numbers stabilize fully 48–72 hours after the month closes.

Daalu subscription

What you’re paying us. Includes:

• Plan name (Starter / Team / Business / Enterprise).
• Seat count (active users).
• Add-ons (local inference, extended retention, etc.).
• Current month-to-date charge.
• Next invoice date.

Download invoice → PDF of the last issued invoice.

Manage plan → Stripe-hosted portal where you change card, update billing email, change plan tier.

If you’re on an Enterprise contract, the panel shows your contracted spend and any overage; invoicing happens out of band.

Spend breakdown and forecast

A combined view spanning cloud + Daalu + LLM:

• By month — bar chart of total spend per month for the last 12.
• By category — pie chart of “compute, storage, transfer, managed services, Daalu, LLM” for the current month.
• By source — same as cloud spend section but unified.

The forecast is a straight-line extrapolation of the current month-to-date. It’s a rough number; for tighter forecasts, hook your accounting team’s tools into the exported CSV (next section).

Exporting

Top-right of the page: Export → CSV / JSON.

The CSV is one row per (cloud-account, service, tag-set, day). Importable into the spreadsheet of your choice.

The JSON has the same data with nested structure preserved. Useful for piping into a warehouse via a workflow.

Both exports respect the current filter selection (date range, selected accounts). Exporting the whole tenant’s history is also possible from Settings → Tenant → Export billing.

Alerts on spend

Cost anomalies fire as alerts (see the cost-anomaly agent in Chapter 19). You can tune the thresholds at Settings → Briefings → Cost anomaly:

• Daily threshold: % deviation from trailing 7-day average.
• Service threshold: per-service spike threshold.
• Notification channel: who gets paged.

Default thresholds catch obvious “your bill is 3× yesterday’s” spikes; tune them tighter once you’ve seen a few false positives.

What’s not on this page

A few things on purpose:

• Detailed cost allocation across teams. Cost-allocation requires consistent tagging across your clouds, which most teams don’t have. The page shows tag-based breakdowns when available but doesn’t try to be a Cloudability replacement.
• Cost optimization recommendations. Things like “rightsize this EC2 instance” are surfaced via the Assistant and the recommendations agent, not on this page.
• Per-user usage. Daalu doesn’t bill per-action, so per-user cost is rarely meaningful. The plan is per-seat.

What’s next

Chapter 23 covers the Reports page — AI-generated briefings plus a query tab for ad-hoc and saved analyses.