# Base44 AI Playbook: Prompts, Models and Credits That Last

> How Base44's AI builds apps, which models it runs, and how to prompt so message credits last: Discuss mode, model choice on Builder, branches and testing.

_Source: https://professionalstoolkit.com/articles/base44-ai-playbook — The Professional's Toolkit · updated November 2026_

---

Base44 is a vibe-coding tool: you describe an app in plain English and its AI chat writes the frontend, database, logins and hosting for you. As of November 2026 the chat runs on Auto by default; from Builder up, the picker also offers Base 1, Base44's in-house model, and models from Anthropic, OpenAI and Google. Picking the model yourself starts on the Builder plan, $40 per month billed annually or $50 per month billed monthly. The real skill is budgeting. **Every change costs roughly 0.5 to 4 message credits** by Base44's estimates, so a good prompting routine decides how far a month's allowance stretches.


**~0.3** — Message credits per Discuss mode message, against about 0.5 to 4 for a change in Build mode (Base44's estimates).



- **~1 credit** — First build of an app, however detailed the prompt
- **Builder+** — Plan needed to pick the AI model or effort level
- **25 a month** — Free plan message credits, at most 5 a day, personal workspace only


## The models under the hood, and who gets to pick

On Free and Starter you get Auto and nothing else; other models show a Builder+ badge in the picker. From Builder up, the Tools icon lists Auto, Base 1, Opus 5.5, Sonnet 5.5, GPT-6.1 Sol, Gemini 4 Argon, GPT-6.1 Terra, GPT 6 Astra, Gemini 3.8 Flash, Fable 5.1 and Haiku 5.5, with some still rolling out gradually. Base44 announced Base 1 in June 2026 and says it keeps using outside models alongside it. Enterprise admins can switch off whole providers or block high-credit models for their members.

Choosing is not free. A model you pick by hand may use more credits than Auto, and Base44 shows the cost of each message under Credits Used once it finishes.[1] Builder also unlocks effort levels from Auto through Low, Medium, High, Extra high and Max; higher effort takes longer and burns more. Compare models, also on the Builder plan or higher, sends one message to two models side by side at double the credits, so a two-credit module becomes roughly a four-credit experiment. Use it once, when a stubborn bug keeps returning, not as a habit.

Most builders should stay on Auto. It is the only option below Builder anyway, and a hand-picked model may cost more credits per message than it does. Reach for a named model when Auto keeps misreading the same instruction twice in a row, then return to Auto once the problem is solved. Rewording the request in Discuss first is usually the cheaper experiment, because a clearer instruction often fixes what a stronger model would only paper over.



> ⚠ **Failed output is still billed:** Credits spent on failed or unsatisfactory AI output are not refunded; Version History undoes the change, not the charge.

## Load everything into the first prompt

The opening build is the bargain of the whole project. Base44's help says **the first build of an app uses about 1 credit**, regardless of how detailed the prompt is or which model runs it. Later changes are priced by how much work they involve. That asymmetry rewards a long, specific first message: name every page, the data each table holds, who logs in and what each role may see, plus the look you want.

You can attach images, PDFs, documents, spreadsheets and slide decks to that first prompt as context. On the Builder plan or higher you can also connect tools such as Google Workspace, Slack or GitHub from that first prompt, before building starts. Start from URL copies the layout and visible content of a public page on any plan, though it skips backend logic, private integrations and hidden data, and it does spend credits while generating. Be patient here. A detailed first prompt can take 10 to 15 minutes, and the chat only warns that it might be stuck after 10.

Plan mode is the warm-up. It asks one or two compact cards of up to three questions each, costs a fraction of a credit per message and holds up to 50 messages before you press Start Building. Answer those cards fully; every gap you leave becomes a paid fix later. Use the remaining room in that conversation to settle the data structure and user roles, since those are the changes that ripple across every page once the app exists.


> 💡 **Front-load the opening prompt:** The first build costs about 1 message credit whatever the prompt's length, so describe every page, data table, role and style before you send it.


## Talk in Discuss, change in Build, nudge by hand

After the first build, the chat has two modes, toggled with Ctrl+. or Cmd+. Build acts at once. Discuss plans and answers questions without touching code until you approve, and it runs on its own cheaper model whatever you selected, so each message typically costs about 0.3 message credits. A third surface, Edit mode, handles visual work: dragging elements and retyping text yourself costs nothing, while asking the AI to rework a single element through Edit Element is billed by size and complexity.

| Action in the editor | Message credits (Base44's estimates) |
|---|---|
| First build of a new app | about 1 |
| Simple text or visual change | about 0.5 |
| Small feature | about 1 |
| Complex module | about 2 |
| App-wide change | about 3 to 4 |
| Discuss mode message | about 0.3 |
| Manual drag or text edit in Edit mode | 0 |

By our arithmetic, ten Discuss messages cost about 3 credits, roughly the price of one app-wide change. That is the trade worth making. Talk a big change through in Discuss, let it propose the steps, then send each step to Build as a small, specific request. A 100-credit month covers about 200 simple edits at 0.5 each, or only about 25 app-wide rewrites at 4 each.[2]

On Free the math is tighter. Free gives 25 message credits a month, at most 5 a day, and only your personal workspace gets them. One app-wide change can eat most of a day's 5, while the same 5 credits buy about 16 Discuss messages. The daily limit resets at midnight in the owner's timezone or UTC, so a long change is best sent at the start of a fresh day. Free users should plan heavily and build sparingly; the [Free plan guide](/articles/is-base44-free) covers the rest of its limits.

## Undo, branch and test before you pay twice

Mistakes are where credits quietly vanish. **Credits spent on failed or unsatisfactory AI output are not refunded**, and Base44 points to Version History as the only undo. Every AI change is saved as a version you can revert to, and you can even publish an older version to the live app while the broken draft stays open for repair. Revert first, then rephrase. Asking the AI to fix its own wrong turn often costs more than rolling back.

Branches let you try two ideas in parallel, each with its own chat and preview, then merge the winner into main, where publishing always happens. There is no cap on branches, up to 20 builds can run at once across main and branches, and branch work draws on the same credit balance as main.[3] One catch matters more than the rest: a branch works on the app's live data, so a record changed in a branch preview changes for every user.


> ⚠ **Branches touch live data:** Records added, changed or deleted in a branch preview change the same data the app uses everywhere; a separate test database needs the Builder plan or higher.


For quality checks, three tools cost little or nothing. Verification reviews code in the background as the builder writes it. Resolve with AI fixes errors flagged in the preview without using credits, and automatic AI fixes are free on paid plans but charged on Free. The Testing Agent, on every plan, walks through the finished app like a new visitor and uses around 1 to 4 credits per run.[4] It cannot test login forms, one-time codes or email verification links, so check those yourself.

## When to stop prompting and open the code

Some fixes are cheaper by hand. The Code tab edits any file with a split-screen live preview; Free can only view code, and in-app code edits start on Starter. If you know which line sets a color or a label, change it there instead of paying the AI to find it. Base44 also lets you freeze files or entities so the AI never modifies them, which protects a hand-tuned page from the next app-wide request.

On Builder and higher, skills store reusable instructions such as brand voice or coding style, up to 50,000 characters each, with up to 100 per workspace. Base44 is candid about their reach: they are static text that steers the editor chat only and cannot pull live data or enforce code standards. Builder also adds Send to Coding Agent, which hands the app to an external agent such as Claude, ChatGPT or a local CLI agent, plus GitHub sync for developers who prefer their own editor.

Design work follows the same logic. A redesign prompt returns four options side by side, a section-level redesign leaves the rest of the build alone, and design prompts use message credits like any request. Design systems, from Builder up, cost no credits to create, but applying one to an existing app rebuilds it rather than restyling in place, so apply it early in a project.


> 💡 **Freeze what you tuned by hand:** Freezing files or entities stops the AI from rewriting them during a later app-wide request; in-app code edits start on the Starter plan.


## Small habits that protect the monthly allowance

A few behaviors decide whether credits last to the reset. Up to 7 messages can wait in the queue while the AI works, so line up small, related requests instead of one sprawling paragraph. Stopping the AI before its first reply is free; stopping mid-response still charges for replies already finished. Web search inside the chat uses credits with any model, and MCP connections, available on the Builder plan or higher, can raise usage too. Each component from the library is one build message, so add components one at a time.

Watch the Credit usage page, which breaks spending down by member, app and source. Unused message credits expire at each monthly reset and do not roll over (the only exception is a one-cycle carryover when you upgrade from a monthly plan to a higher monthly plan), and Base44 does not publish a way to buy extra credits outside a plan, so a heavy month ends in a wait or an upgrade. The [pricing guide](/articles/base44-pricing) shows what each plan's allowance buys before you pick one.


## FAQ

**Does Base44 use my prompts to train AI models?**

On every plan except Enterprise, Base44 can use workspace data to train AI models and there is no opt-out setting. Only Enterprise workspaces are excluded, automatically.

**Can I build with Base44's AI from my phone?**

Yes, from a phone browser or the Base44 app for iOS and Android. Edit mode, data permissions, code editing, APIs and app templates are desktop only.

**Can I hand my Base44 app to Claude Code or Cursor?**

Send to Coding Agent, available on the Builder plan or higher, passes the app to an external agent. Base44 also publishes skills that teach coding agents such as Claude Code and Cursor how to work with Base44 projects through its CLI, which is in beta.

**What happens if I run out of message credits mid-project?**

You can still type a message, but sending it needs credits, so AI building pauses until your credits reset: at midnight if a Free user only hit the 5-a-day cap, otherwise at the next monthly reset. Base44 does not publish a way to buy extra credits outside a plan; upgrading is the documented route.


## The bottom line

Base44's AI rewards planning more than clever wording. Pack the first prompt, since it costs about one credit whatever its length. Think aloud in Discuss at about 0.3 a message, then build in small, named steps. Do visual tweaks by hand for free, and roll back with Version History instead of arguing with a bad change. Stay on Auto unless a specific problem justifies Builder's model picker, where a hand-picked model may cost more credits per message. Builders who follow that routine get far more working app out of the same monthly allowance.



## Sources

[1] Base44 docs: Using the AI Chat in Base44 — https://docs.base44.com/Building-your-app/AI-chat-modes (2026-11)
[2] Base44 docs: Credits — https://docs.base44.com/Account-and-billing/Credits (2026-11)
[3] Base44 docs: Working with branches — https://docs.base44.com/Building-your-app/working-with-branches (2026-11)
[4] Base44 docs: Testing your app's flows — https://docs.base44.com/documentation/managing-app-data/testing-agent (2026-11)
[5] Base44 pricing page — https://base44.com/pricing
[6] Base44 Terms of Service — https://base44.com/about-us/legal/terms-of-service
[7] Base44 Privacy Policy — https://base44.com/about-us/legal/privacy-policy