Office Hours #63 · Open drop-in · a practical reset
A practical Q&A on billing changes, missing translations, branch limitations, large data jobs, and app-store publishing.
Understand the task, check usage, and bring evidence to support when a run looks wrong.
Session #63 / The recap
Sam and Robert hosted an open question session after Basecamp's public announcement. The recording begins after Robert had already covered the opening portion on Discord, and Sam joined late because the launch communications coincided with Office Hours.
The main thread was getting builders unstuck: an annual-to-monthly billing change, credit-heavy iteration loops, explicit instructions for installed packages, and the distinction between app-level API integrations and the Base44 builder. The hosts also discussed branch limitations, a 50,000-record text-processing job, launch visibility, and why app-store submission still needs the developer's involvement.
The useful habit
A reproducible example beats a guess. The session covered account questions, code workflows, integrations and launch blockers—with limits made explicit.
Every substantive question
Search all 24 answers, or filter by topic. Open a question for the full response.
24 of 24 answers
Sam said dedicated one-to-one staff help was not offered in that format. He recommended community help, voice channels, or Office Hours, and asked users to report unclear testing UX. The Stripe test setup itself was not walked through in the recording.
No. Robert had already been present on the hour, and the session would finish at the normal time, also allowing the next Spanish session to begin.
The hosts suspected that the package had been installed without fully integrating it, or the builder had manually translated only part of the app. Explicitly request use of the installed package across all pages and specify the selector and languages. Their one-prompt claim was not a verified guarantee of complete or accurate translations.
The hosts explained that it may require escalation, recalculation of the annual discount, confirmation of changed entitlements, and manual checks. They agreed that the reported 48-hour delay was poor and offered to escalate the ticket; they did not complete the change live.
A change can affect price, domains, backend functions, and other entitlements. The hosts said support seeks explicit confirmation of the consequences and may need to check multiple production apps rather than treating it as a simple toggle.
Sam and Robert offered to review the ticket number and route the unusual case to the billing team and support management. That was an offer to investigate, not proof that the issue was resolved or a promise of a faster SLA for every ticket.
The hosts mentioned qualifying event entries, referrals, and achievements or milestones. Build for Good was offering 20 credits for an approved legitimate submission at that time. Conditions still apply; the recap does not present these as unlimited guaranteed grants.
Robert checked the settings and read out English, Japanese, Dutch, Spanish, French, and Portuguese. This was separate from adding translations to a user's own app.
Identify the specific problem and desired visual direction, attach design information, inspect relevant examples, and test the result. Robert warned that gratuitous effects can distract from the app's actual purpose even if they look impressive.
The hosts described packages as dependencies that add reusable capabilities and pointed to the dedicated earlier npm Office Hours session. They emphasised that installing a package is only the first step; the app must actually use it.
Sam said branch backend-function support was not yet part of the core implementation at the time and was coming. The session did not demonstrate a workaround or confirm when it would ship.
The hosts had seen possible stalling and time delays, including a mismatch between user-visible messages and their own checks. They did not diagnose the specific report or provide a complete branch compatibility list. Code-view editing trouble was also raised without a confirmed fix.
Usage depends on the model, complexity, and repeated attempts. The hosts recommended Discuss, Plan, or visual edit mode as appropriate and avoiding expensive Fable runs when budget matters. They did not inspect the user's usage ledger.
Sam's method was to revert before the loop, reframe the task, and try a different model. A voice conversation with an assistant can clarify requirements and generate a better prompt, but may itself incur usage.
The hosts avoided a universal recommendation. They suggested trying the platform and upgrading as the required features and usage become clearer, with the choice depending on commitment, finances, and expected needs.
No, according to the hosts. App-level calls to an external model API are a separate integration path; they do not replace builder prompt credits. API billing should be checked separately from any consumer ChatGPT subscription.
The hosts said they are meant to be aspirational and encourage activities such as integrations and sharing. Some milestones could be redeemed for credits, so meaningful requirements were intentional.
The hosts expected a normal upgrade to update promptly, but the posted evidence suggested a possible manually handled billing case. They could not diagnose it from chat and requested a ticket number for escalation.
Sam recommended exporting the entity, performing a programmatic transformation outside the platform with a tool such as Codex or Claude Code, and then importing the results. No live batch or successful re-import was shown; back up and validate data before applying such a workflow.
No. The hosts said the same agent can switch languages while retaining conversation context. If a specific default language is required, state it in the instructions; they also discussed similar multilingual behaviour in in-app agents.
Sam thought the process could be streamlined but not realistically reduced to that promise: Apple and Google still require business or developer information, payment, and identity checks. He spoke from experience and did not announce a new publishing service.
Use the guide in the publishing flow and open a support ticket identifying the exact confusing step. Sam said this feedback helps improve the walkthrough; he did not see a separate downloadable PDF as inherently necessary.
Share with the actual communities that benefit, plus peers and personal networks where relevant. Votes improve visibility, but Sam said staff review entries and a panel applies criteria; the most-voted app does not automatically win.
Robert recommended feedback.base44.com to find existing requests or add a clear proposal. Support can record usability issues, but the feedback site was described as the more direct product-feedback route.
What was taught
6 teaching points, with the practical details and limits preserved.
After approving an npm dependency, explicitly ask the builder to use it across the relevant feature. Sam's translation example requested the installed internationalisation package, complete page coverage, named languages, and a top-right language selector. Installation alone does not supply or validate every translated string.
Sam reverts to before the first prompt that began an unproductive loop, then changes the approach and sometimes the model. Another option is to describe the desired result by voice to an assistant, let it ask clarifying questions, and turn that into a clearer builder prompt.
For a simple visual change, visual editing may be more direct than another broad build prompt. Discuss and Plan modes help clarify an unresolved problem before further edits. See Base44 AI chat modes (opens in a new tab). Model choice and task complexity affect credit consumption.
A custom app can call an external LLM API using configured credentials instead of the built-in InvokeLLM path. That is different from replacing the Base44 builder's model billing with your own tokens, which the hosts said was unsupported. Consumer chat subscriptions should not be assumed to include API usage.
For 50,000 text records, Sam suggested exporting the entity, processing it programmatically with a coding tool outside the app, then importing the results. A safe implementation should preserve a backup and record IDs, validate a sample, and verify the import rather than treating a large overwrite as risk-free.
Robert suggested sharing Build for Good entries with communities that genuinely benefit. This produces more relevant feedback and visibility than only asking other builders for votes. Sam stressed that a human panel still evaluates submissions on criteria.
Keep these close
From the session
These updates reflect what was said on 26 August 2026, not a promise of current availability.
Basecamp had been available for roughly 24 hours and was now announced across company social channels. Sam credited Monica, Sam, and Tiffany and said iterative improvements would follow.
Selected users were testing ad-hoc message and integration credit purchases, including annual-plan users. The hosts saw this as useful for people who run out of credits without wanting a full plan change; universal access was not announced.
A legitimate approved entry could receive 20 credits. The hosts reminded viewers that staff judging and criteria matter, not just popularity or raw vote totals.
Sam said backend functions in a branch were not yet core supported functionality and were coming. He also acknowledged reports of sync delays. No delivery date was given.
A Spanish Office Hours run by Jordi was due immediately after the session. Yoav was announced for the following day to discuss a new product release; Twitch expansion was also mentioned without a launch date.
The original recording
Watch the demonstrations, questions and in-the-moment decisions in full.
Continue from here
Named resources are retained without inventing missing links.