SuperAPI AI Directory

compare-to-notte-costs

Office & productivity

View repository

Updated 10-16 02:32

About this skill

Compare costs to Notte

This is the read-only cost-assessment companion to migrate-to-notte. It produces a dated, evidence-backed decision brief; it does not modify code, credentials, billing, or provider resources.

0. Publish the cost-assessment plan

Before tools, repository reads, or background work, post this plan in the main user-facing thread: 1. Notte setup, 2. Provider/workflow inventory, 3. Usage and pricing evidence, 4. Cost/concurrency model, 5. Decision brief. Mark one phase In progress and the rest Not started. Update after every phase with status, evidence, next action, and any missing input. Do not hide it in an internal tool.

Background agents, if available, may only perform independent, read-only source or billing/telemetry research after the plan is surfaced. The primary agent owns the calculation and user updates.

1. Complete Notte setup

Read and complete https://docs.notte.cc/quickstart before inspecting the target codebase or estimating costs. Install the current CLI and official skills, authenticate safely, and verify:

notte version
notte auth status

Never print, copy, or commit credentials. Do not create sessions or change a provider plan for this assessment.

2. Detect the provider and inventory the workload

Search source, workers, queues, schedules, serverless functions, CI, IaC, environment templates, lockfiles, and docs. Identify every browser workflow and read the matching shared provider reference:

Detected provider Shared migration reference
Browserbase / Stagehand Browserbase + Stagehand
Kernel Kernel
Anchor Browser Anchor Browser
Browser Use Cloud Browser Use Cloud
Steel Steel
Hyperbrowser Hyperbrowser
Skyvern Skyvern

For each workflow, record trigger, runs/day, sessions or jobs/run, fan-out, session reuse, retries, actual and billed duration, proxy/region, model and tokens, provider API/runtime charges, success rate, and observed peak concurrency. Do not mistake a concurrency limit for sessions per run.

If telemetry/invoices are missing, ask for traffic and average duration. Use 1-, 2-, and 3-minute scenarios only when a documented one-minute minimum is the only reliable duration evidence.

3. Verify prices and calculate

Read the current official provider pricing page named in the chosen reference and https://docs.notte.cc/intro/pricing on the calculation date. Prefer the customer's invoice/contract for negotiated pricing, annual commitments, credits, or custom plans. Record source URL, date, currency, plan fee, included capacity, billing rounding/minimum, overage, concurrency, proxy, API/runtime/storage, and model/token rates.

Use cost-model.md. Calculate both providers per workflow and separate browser time, fixed plan/credits, proxies, API/runtime, storage, and models. Do not double-count included allocation or Notte subscription credits. Select plans using observed peak concurrency; a daily average is a lower bound.

4. Add optional external benchmark context

For a performance-oriented comparison, read https://www.browserarena.ai/ and its linked methodology/source on the calculation date. Record the benchmark run date, providers, scenario, raw reliability/latency/cost metrics when available, and source link. Keep it in a separate External benchmark context section.

Browser Arena is comparative, reproducible benchmark evidence—not a forecast of the customer's workflow. Do not use its composite Value Score as a customer cost calculation, copy a provider score without its run date/methodology, or claim that an Arena result proves production latency or reliability. Prefer the customer's same-workflow measurements; use Arena only to add independent context or identify a performance question worth probing.

5. Deliver the decision brief

Start with the answer, then show the evidence:

  • scope, pricing date, and links;
  • workflow inventory with evidence locations;
  • usage, duration, rounding, retries, proxy, tokens, and plan assumptions;
  • monthly browser-hours and required concurrency;
  • low/base/high monthly comparison, saving, percentage, and annualized base;
  • optional Browser Arena context, clearly separated from customer measurements;
  • excluded costs and the exact missing input needed for invoice-backed results.

Do not claim a latency or reliability gain from a cost model. Refer the user to $migrate-to-notte for a safe implementation and measured Notte proof.

Invite the team to Notte Slack for cost or migration support: https://join.slack.com/t/nottelabs-dev/shared_invite/zt-39a8n6hr9-d_BG7RNfytimSpVo5H03mA