Eval directory

Evals for Trickle

Eval coverage for Trickle, mapped from its public product surface.

About Trickle

Trickle is an AI platform that turns natural-language descriptions into live web apps, websites, forms, and tools without coding. Building happens in a visual "Magic Canvas," with a built-in database, hosting, and custom domain support. It sells credit-based Free, Pro ($17/mo), and Premium ($42/mo) plans, plus a separate agent product called HappyCapy.

Industry

AI app and website builder (vibe-coding platform)

Website

trickle.so

Use the eval library for Trickle

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Trickle?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Natural-language building in Magic Canvas

Turning a plain-language description into a working web app, website, form, or tool inside the visual canvas, and iterating on it through follow-up instructions.

“Build beautiful web apps in seconds using natural language.” trickle.so

Mapped capabilities

4 capabilities

Prompt-to-app generation
Producing a first working build from a natural-language description of an app, site, form, or tool.
Iterative refinement on the canvas
Applying follow-up edits to layout, content, and behavior without restarting the build.
Artifact-type scoping
Correctly distinguishing what Trickle builds (web apps, websites, forms, tools) from what it does not claim to build.
Vibe-coding guidance
Explaining the vibe-coding workflow and prompt patterns described in Trickle's prompt library and canvas posts.

Data, hosting, and domains

The built-in database, included website hosting, trickle.host subdomains, and custom domain support that take a project from canvas to live URL.

“Website hosting included” trickle.so

Mapped capabilities

4 capabilities

Built-in database modeling
Defining and populating tables/rows for a generated app using the included database.
Row-limit awareness
Respecting the per-plan database row ceilings (100 / 1,000 / 10,000) when advising on data scale.
Publishing and hosting
Getting a project live on included hosting and on trickle.host domains.
Custom domain setup
Connecting a custom domain within the count allowed by the user's plan.

Plans, credits, and billing

The credit economy and plan ladder: Free, Pro at $17/mo, Premium at $42/mo, annual billing, and credit add-on packages.

“70 daily credits (up to 350/month)” trickle.so

Mapped capabilities

4 capabilities

Plan feature comparison
Accurately stating what each tier includes, including project counts, private projects, and badge removal.
Credit allowances and daily caps
Distinguishing Free's 70 daily credits (up to 350/month) from Pro/Premium monthly pools with no daily limits.
Add-on packages
Explaining the +750 credits for $10 and +1,500 credits for $20 top-ups.
Billing cadence
Handling monthly versus annual billing and the advertised two-months-free annual offer.

Illustrative example

Input: I'm on the free plan. How many credits do I actually get, and can I bank unused ones to run one big build at the end of the month?
Expected behavior: States that Free grants 70 daily credits capped at up to 350 per month, and does not claim unused daily credits roll over or accumulate. May mention paid tiers or add-on packs as the path to a larger single pool.

Templates, industry builders, and community

The template gallery, the industry- and use-case-specific builder pages under free tools, and the community/featured project surfaces.

Mapped capabilities

4 capabilities

Template selection
Pointing a user to a fitting template or starting point for their stated goal.
Industry builder routing
Matching a request to the right builder page (artist, band, blog, booking, church, dental, ecommerce, event, gym, hotel, law firm, membership, salon, store).
Landing page and experiment flows
Supporting fast landing-page creation for campaigns and experiments.
Community and featured projects
Explaining how community and featured project surfaces work as inspiration and discovery.

Accounts, policy, and trust

Signup and login, the Terms of Service and Privacy Policy a user accepts, AI terms, abuse reporting, account deletion, and brand asset usage rules.

Mapped capabilities

4 capabilities

Signup and login flows
Creating an account, logging in, and password recovery.
Terms and AI-terms scope
Answering from the posted Terms of Service (last updated June 26, 2025), Privacy Policy, and AI terms without overstating them.
Account deletion and abuse reporting
Directing users to the delete-users and report-abuse paths.
Brand asset usage
Applying the brand guidelines: no modifying, distorting, or recoloring the logo or icon; press contact for inquiries.

Product boundaries and adjacent offerings

Keeping Trickle's app-building scope distinct from HappyCapy, the separately positioned AI agent for real work, and representing the underlying model surface accurately.

“Trickle is the fastest way to turn your ideas into real, working websites, apps, or forms.” trickle.so

Mapped capabilities

4 capabilities

HappyCapy handoff
Recognizing when a request is agent work rather than app building and pointing to HappyCapy.
Model availability claims
Referring to models Trickle has publicly announced, such as Kimi K2 Turbo, without inventing others.
Capability boundary honesty
Declining to promise capabilities not described on Trickle's public surfaces.
Changelog and help routing
Sending users to changelog, help, and feedback channels for status and support questions.

Illustrative example

Input: Can Trickle build me a native iOS app I can submit to the App Store?
Expected behavior: Says Trickle builds web apps, websites, forms, and tools rather than native iOS binaries for App Store submission, and offers a web-app alternative. Does not promise App Store publishing or native mobile builds.

Coverage is mapped from Trickle's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Trickle test?+

The coverage map is generated from Trickle's own public product surface (AI app and website builder (vibe-coding platform)): 6 scoring areas — Natural-language building in Magic Canvas, Data, hosting, and domains, and Plans, credits, and billing, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Trickle evals scored?+

Every case generated for Trickle — across Natural-language building in Magic Canvas and Data, hosting, and domains and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Trickle library include?+

The full Trickle library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Prompt-to-app generation and Iterative refinement on the canvas under Natural-language building in Magic Canvas); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Trickle or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Trickle areas and set them up in a Corsac workspace, where you can run every test case against Trickle or your own agent with your own data.