Home / Add-ons / Guided Onboarding
Guided onboarding — from API key to a working integration in two sessions
Two live sessions with a Plugsky engineer: keys, first integration, an evaluation set and a cost baseline. Delivered in one week.
Who it is for
Teams that have the API key and the docs but want the first integration reviewed by someone who has done it before.
The problem it removes
- The first integration works in a notebook and breaks in production.
- Nobody knows which model fits the workload.
- The cost baseline is a guess until the first invoice arrives.
What you get
- Session 1: keys, auth, streaming, error handling and rate limits
- Session 2: model choice for your workload, with an evaluation set
- A written cost baseline using your own volumes
- Recording and notes after each session
- 30-day follow-up by email
How to use it
1. Send the intake form
Tell us the stack, the workload and the deadline. We confirm scope before booking.
2. Two live sessions
Screen-shared, recorded, with your code on screen — not a slide deck.
3. Leave with a baseline
A written summary: models chosen, limits to watch, and the monthly cost at your volume.
What the two sessions actually cover
Each session is 60 minutes, screen-shared, recorded, and run against your repository rather than a sample project. The agenda is fixed so the time is not spent on introductions, but every example is yours. If your integration is in a language we have not seen before, that is fine — the review is about what the request sends and what comes back, not about the framework.
| Segment | Time | What happens | What you leave with |
|---|---|---|---|
| Session 1 — connection | 60 min | Keys and environment variables, base URL, streaming, timeouts, retry and error handling, reading rate-limit headers | A request that works from your machine and from staging |
| Session 2 — evaluation | 60 min | A 20–50 case evaluation set built from your real inputs; two or three models compared on the same cases | A scored comparison and one model chosen with a written reason |
| Written baseline | Within 24 h | Your monthly volume applied to the published limits | A one-page cost baseline naming the cap you would hit first |
The order matters. Most teams pick a model before they have an evaluation set, then discover in production that the cheaper model fails on the one input class that matters. Session 2 exists to prevent that. The evaluation set is small on purpose: twenty cases you can read in full beat two thousand cases nobody has inspected.
What we need from you before the first session
The sessions are only as good as the inputs. Five things are needed, and four of them you already have.
- An API key you can use during the call, or the ability to create one.
- Access to the integration code — a screen share is enough if you cannot grant repository access.
- Twenty to fifty real inputs you would send in production. Redacted is fine. Synthetic inputs hide the edge cases, which is the opposite of what an evaluation is for.
- Expected month-one volume: requests per day, average tokens per request, peak concurrent calls.
- One binding constraint — latency, cost or output quality. Pick one. The recommendation changes depending on which you name, and a session that tries to optimise all three produces advice that optimises none.
If you cannot supply real inputs, the session still runs, but the evaluation is weaker and the written summary says so. We do not present a synthetic benchmark as if it were your workload.
What a cost baseline contains
The baseline is arithmetic on your numbers, not a promise about your future bill. It has five lines, and the fifth is the one people are surprised by: the constraint that binds first is usually not the one they were watching.
| Line | Source | Example |
|---|---|---|
| Requests per day | Your logs, or a stated estimate | 1,400 |
| Tokens per request | Measured on 20 sample calls | 2,800 |
| Tokens per day | Product of the two | 3,920,000 |
| Cap that binds first | Published plan limits | Starter token cap — 5,000,000/day |
| Monthly price | Published plan price | $20 |
In the example, Hobby's 1,000 requests per day would have failed before the token cap was ever reached. That ordering matters, because teams watch tokens and are surprised by the request limit. The baseline checks both, in the order the API enforces them.
What the written summary looks like
You receive one document, not a slide deck. It is written to be forwarded to whoever approves the bill, so it contains the decisions and the numbers behind them, and nothing that only makes sense if you were on the call.
- Stack summary — language, framework, where the request is constructed, where the response is parsed.
- Configuration findings — anything in the request that costs more than it needs to: missing timeouts, unbounded context, retries without jitter, streaming disabled.
- Model comparison — the evaluation cases, the scores, and the reason for the choice.
- Cost baseline — the five lines above, with the source for each.
- Watch list — the three limits most likely to be hit first, and what to do when they are.
- Recording — a link to both sessions, valid for 90 days.
What the sessions do not include
Onboarding is advice and review, not implementation. We do not write your production code, deploy to your infrastructure, or take access to your production database. If you want the integration written for you, that is a separate engagement and we will say so at intake rather than quietly expanding the scope halfway through.
- No production access. Staging or a local environment is enough for everything in the agenda.
- No code committed to your repository. You get patch guidance and a recording; your team applies it.
- No guarantee about a third-party provider's behaviour. If you are moving between providers, the migration service covers that ground.
- No legal or security certification. If a security questionnaire is the actual blocker, say so at intake and we will tell you what we can and cannot sign.
How long it takes
| When | What happens |
|---|---|
| Day 0 | Intake form received; scope confirmed by email |
| Day 1–3 | Session 1 booked at a time you choose |
| Day 3–7 | Session 2, after you have run the evaluation set |
| Within 24 h of session 2 | Written baseline and both recordings delivered |
| Day 30 | Follow-up by email: what changed, what broke, what to watch |
The week is the normal case. If your team cannot run the evaluation between sessions, the second session waits and the booking does not expire.
Who should skip it
If you have already shipped an integration and you have a working evaluation set, $150 buys very little. The same is true if your workload is a single prompt with no model choice to make, or if the only question is price and the pricing page already answers it. In those cases the documentation is enough, and we would rather say so than take the booking.
Questions
What does it cost?
$150 one-time for two sessions and the written summary. No subscription.
Who runs the sessions?
A Plugsky engineer, not a salesperson. You can bring your own code.
How soon can we start?
Usually within three business days of the intake form.
Is it refundable?
Scope and refund terms are confirmed in writing before booking. If the first session does not cover the agreed scope, tell us within 24 hours and we will agree a re-run or a refund in writing.
Does it include implementation?
No. We review and advise; your team writes the production code.
Sources
- Plugsky pricing — plan limits and fair use — checked 21 November 2026
- OpenAI chat completions reference — checked 21 November 2026
- Stripe subscriptions overview — checked 21 November 2026
Related add-ons
Migration Service
A compatibility review, a shadow-traffic run and a quality diff before cutover, so you change the base URL with evidence instead of hope.
Compliance Audit
A scoped technical assessment: what applies to you, what is missing, and the order to fix it. A report and an evidence appendix, not legal advice.
One intake form, then two sessions in the same week. Paid once through Stripe; nothing renews.
Book the engagementCompare add-ons