What an autopilot SEO run actually costs.
Every API call a full SEO/GEO loop makes, priced at published rates and sourced to the line of code that bills it. You pay the providers directly — we take no margin on any of it.
Where the money goes.
One day of the Core loop — 10 pages drafted, edited, QA’d and published, plus 100 tracked queries and grounded visibility probes across the LLMs.
Drafting, editing, QA verdicts and grounded visibility probes. Billed by Anthropic against your key.
Brave Search calls for keyword expansion, rank checks and citation checks. Billed by Brave against your key.
Nominal per-publish entry so the ledger reflects the work. Not a vendor invoice.
No worker bills this category today. Shown at zero rather than estimated.
No worker bills this category today. Shown at zero rather than estimated.
One full loop, every call.
In the order the scheduler actually runs them. The source column is the constant in the codebase that produces each rate — check our work.
| Step & source | Volume | Unit rate | Cost |
|---|---|---|---|
Keyword + SERP expansion researcher · SERP_PAISE — src/workers/researcher.ts:27 | 5search call | ₹0.30 | ₹1.50 |
Draft a page writer · claude-sonnet-4-6 @ published rates — src/llm/client.ts:28 | 10page drafted | ₹5.11 | ₹51.10 |
Edit + tighten the draft editor · claude-sonnet-4-6 @ published rates — src/llm/client.ts:28 | 10page edited | ₹5.36 | ₹53.60 |
Quality gate (pass/fail) qa · claude-haiku-4-5, 8-token verdict — src/workers/qa.ts:80 | 10page checked | ₹0.34 | ₹3.40 |
Publish + serve the page publisher · PUBLISH_PAISE — src/workers/publisher.ts:17 | 10page published | ₹0.05 | ₹0.50 |
Rank + citation checks monitor · SERP_PAISE — src/workers/monitor.ts:42 | 100tracked query | ₹0.30 | ₹30.00 |
Grounded LLM visibility probes monitor · PROBE_PAISE, capped by dailyProbeCap() — src/workers/monitor.ts:43 | 20probe | ₹0.20 | ₹4.00 |
| One full loop — core tier, per day | ₹144.10 |
Vendor spend: ₹143.60 of ₹144.10 is money leaving your account to Anthropic and Brave. The rest are nominal ledger entries for hosting and embeddings.
Assumptions: ~3–4k input and 3.5k output tokens per page, no prompt-cache credit claimed, 30 index chunks per page, and a fixed ₹83 = $1. Real spend trends below this.
The same loop, at every size.
Higher tiers publish more pages and escalate to stronger models, so cost scales with both volume and model choice.
- 3 pages/day published
- 30 queries tracked
- Drafting on claude-sonnet-4-6
- 10 pages/day published
- 100 queries tracked
- Drafting on claude-sonnet-4-6
- 30 pages/day published
- 500 queries tracked
- Drafting on claude-opus-4-7
Per-seat pricing hides this table. BYOK can’t.
A tool that resells you API capacity has every reason to keep these numbers private, because the gap between what the model costs and what you are charged is the business model. Publish the table and you publish your own margin.
Growzo doesn’t have that gap. You bring your own keys, so Anthropic and Brave bill you directly at their list prices. We never touch the money — there is no spread on tokens, no markup on search calls, and no per-seat multiplier on any of it.
That means the ledger inside your dashboard is not our estimate of your spend, it is your spend — the same figures you can reconcile against your provider’s billing page, to the paise.
It also means these costs fall when the providers cut prices, on the day they cut them, without waiting for us to re-tier a plan. And if you self-host the open-source stack, the platform cost goes to zero and this table is your entire bill.
Providers bill your key directly at list price. Nothing sits in between.
Each line cites the file and line that bills it. Our tests fail if they drift.
The loop throttles before it can exceed your monthly budget.
How we built this page
Is this measured spend from real customer accounts?
No, and deliberately so. The cost ledger is project-scoped — every row belongs to a tenant, and there is no marker separating our own usage from a customer’s. Publishing an aggregate over that table would publish our customers’ spend and business volume. So this page reads nothing from the database. It is arithmetic over the same rate constants the runtime ledger bills against, applied to the volumes your tier already guarantees.
Then how do I know the numbers are real?
Every line cites the file and line number of the rate it uses. LLM figures come from the published Anthropic price table in src/llm/client.ts and are routed through the same routeModel() policy the workers use, so a Pro page really is priced at Opus rates. A test suite asserts these constants stay in sync with the workers — if someone re-prices a worker and forgets this page, those tests fail.
Does Growzo mark up API spend?
No. You bring your own keys, so the provider bills you directly at their list price. There is no reseller margin, no per-token spread, and no seat multiplier — nothing sits between you and the invoice. The ledger records exactly what the provider charged so you can reconcile it against your own billing dashboard.
Why is the model probably higher than what you will actually pay?
Two reasons. Every system prompt is sent with cache_control set, so repeat calls read cached tokens at a fraction of the input price — and this model claims no cache credit at all. It also assumes every worker runs at full daily volume every single day, which is the ceiling, not the average.
What about image generation and crawling?
The ledger has categories for both, but no worker currently bills them, so they show ₹0.00. We would rather show an honest zero than estimate a number we cannot source.
What is a “nominal” line?
Hosting and embeddings are recorded at a token rate so the ledger reflects that work happened, not because a vendor sent an invoice for that exact amount. We label them nominal rather than folding them into the vendor total and overstating precision we do not have.
Know your costs before you commit.
Bring your own keys, set a hard cap, and watch the real ledger fill in from day one.
Open-source · Self-host · BYOK · ₹10k/mo hard cap