Verbix

A full-stack workspace for discovering, structuring, versioning, sharing, and running reusable AI prompts.

Repository · Report a bug · Request a feature · Security policy

---

Table of contents

• About
• Capabilities
• Architecture
• Technology stack
• Getting started
• Usage
• API surface
• Project structure
• Testing
• Deployment
• Contributing
• Security
• License

---

About

Verbix makes reusable prompt work discoverable and manageable. Visitors can browse public prompts by category, tag, model compatibility, and discovery ranking. Authenticated creators can compose prompts, define exact {{variable}} tokens, preserve versions, enhance prompts through a server-side AI adapter, and deploy controlled prompt agents. Administrators moderate prompts, reviews, taxonomy, reports, source policies, and imported candidates.

Prompt improvement does not overwrite source work. Verbix stores the original prompt alongside an editable improved prompt and structured details covering intent, assumptions, missing information, constraints, variables, output format, acceptance criteria, and agent notes.

Capabilities

Area	Implemented capability
Prompt authoring	Creates prompts with title, description, body, modality, visibility, compatible models, and version history.
Variables	Parses exact {{variable}} tokens, validates typed values, compiles live previews, and exports text or JSON.
AI enhancement	Uses server-only free-chatbot provider fallbacks with bounded responses and an editable local fallback.
Discovery	Provides featured, trending, and recent sections with search plus category, tag, model, and sort filters.
Community	Supports saved prompts, ratings and reviews, reports, public creator profiles, and creator metrics.
Agents	Deploys owned prompts as shareable agents with server-side variable compilation and persisted hourly visitor rate limits.
Moderation	Enforces administrator-only controls for prompts, reviews, reports, categories, tags, source policies, imported candidates, and external output references.
Rights-aware imports	Processes reviewer-submitted public URLs from approved domains, applies network and robots.txt safeguards, preserves attribution, and stores media only as external references.

Architecture

flowchart LR
  Browser[Browser] --> Client[React + Tailwind]
  Client -->|typed RPC| API[Express + tRPC]
  API --> Auth[Manus OAuth]
  API --> Data[Drizzle ORM]
  Data --> DB[(MySQL / TiDB-compatible database)]
  API --> AI[Server-only free-chatbot adapter]
  AI --> Providers[Phind · DuckDuckGo · Blackbox]
  API --> Importer[Rights-aware source importer]

Loading

The browser accesses typed procedures through /api/trpc. Server-side procedures enforce authentication, ownership, and administrator roles before writing to the database. The client bundle never calls AI providers directly.

Technology stack

Layer	Technology
Frontend	React ^19.2.1, TypeScript 5.9.3, Vite ^7.1.7, Tailwind CSS ^4.1.14, Wouter ^3.3.5
Backend	Node.js, Express ^4.21.2, tRPC ^11.6.0, Zod ^4.1.12
Data	Drizzle ORM ^0.44.5, mysql2 ^3.15.0, MySQL/TiDB-compatible database
AI and extraction	free-chatbot 1.0.2, Cheerio ^1.2.0
Validation	Vitest ^2.1.4, TypeScript compiler, Vite production build

Getting started

Prerequisites

Install Node.js and pnpm. The repository declares pnpm 10.4.1 in its package-manager metadata and includes a pnpm lockfile.

Variable	Purpose
DATABASE_URL	Connection string for the application database.
VITE_APP_ID	OAuth application identifier.
JWT_SECRET	Session signing secret and credential-vault key input.
OAUTH_SERVER_URL	OAuth service base URL.
OWNER_OPEN_ID	OAuth identity assigned the initial administrator role.
BUILT_IN_FORGE_API_URL	Built-in platform service base URL.
BUILT_IN_FORGE_API_KEY	Server-side built-in platform service credential.
NODE_ENV	development locally or production for a built server.

git clone https://github.com/vincenzo-afk/verbix.git
cd verbix
pnpm install
pnpm dev

Database definitions reside in drizzle/schema.ts. After changing a schema, generate a migration, review the SQL, and use the managed database workflow for the target environment.

pnpm drizzle-kit generate

Usage

Browse and use a prompt

1. Open Discover to search the public library or filter prompts.
2. Open a prompt detail page to inspect its content, variables, creator attribution, and approved external references.
3. Supply values for {{variable}} tokens. Verbix validates and compiles the prompt live.
4. Copy or export the result, save it, or run it through the authenticated AI workflow.

Create and improve a prompt

1. Open Compose and enter the source prompt.
2. Include exact {{variable}} tokens for downstream user input.
3. Request an improvement; the original remains available and the result can be accepted as a new version.
4. Save the draft in Workspace, configure typed variables, and submit it for moderation when it is ready for public discovery.

Import public prompt material

1. Submit a specific public HTTP(S) source URL in Imports.
2. An administrator approves the domain and provides its public terms or reuse-policy URL before fetching.
3. The importer applies address-resolution, redirect, content-type, size, and robots.txt safeguards.
4. Administrators review candidates and external output references, then promote approved candidates to the public library.

Verbix does not bypass authentication, payments, access controls, or robots.txt; it does not copy externally hosted media.

API surface

The application has typed tRPC procedures under /api/trpc, consumed by the React client through @trpc/react-query.

Router	Selected procedures	Access
discovery	list, bySlug, creator, categories, tags, related, recordView	Public
workspace	mine, byId, create, saveVersion, variables, submit, toggleSave, review, report	Authenticated; ownership enforced where applicable
improvement	enhance, accept	Mixed public and authenticated
execution	run, history	Authenticated
agents	mine, create, publicBySlug, invoke	Mixed; creation is ownership-gated
credentials	list, set, remove	Authenticated and user-scoped
importer	mine, submitSource, ingest, review and policy controls	Mixed; review and policy controls are administrator-only
admin	Moderation queue, reports, taxonomy, and moderation mutations	Administrator-only

Project structure

verbix/
├── client/src/             # React pages, components, hooks, and routes
├── drizzle/                # Drizzle schema and generated migrations
├── docs/                   # Importer operations and policy documentation
├── server/_core/           # OAuth, tRPC, storage, and server bootstrap
├── server/routers/         # Domain-specific tRPC procedures
├── server/services/        # AI adapter, credential vault, and importer
├── server/db.ts            # Database persistence helpers
├── shared/                 # Shared constants, types, and errors
└── vitest.config.ts        # Node-based test configuration

Testing

Run the full validation sequence before submitting a pull request:

pnpm test
pnpm check
pnpm build

Vitest covers prompt variables, AI response normalization and fallbacks, credential-vault behavior, importer safeguards, authorization boundaries, and router success paths. The CI workflow runs the same commands for pushes and pull requests targeting main.

Deployment

pnpm build
pnpm start

Set all required variables in the target platform’s protected environment configuration. The Express server selects an available PORT at runtime and requires network access to the configured OAuth service and database.

Contributing

Read CONTRIBUTING.md for the development workflow and CODE_OF_CONDUCT.md for community standards. Do not submit secrets, personal data, fabricated user content, or material that violates source rights.

Security

Read SECURITY.md before reporting a vulnerability. Do not publish sensitive security details in a public issue.

License

Verbix is distributed under the MIT License. Copyright © 2026 BHARANI KUMAR S.

Acknowledgments

Verbix uses React, tRPC, Drizzle ORM, Vite, Tailwind CSS, and free-chatbot.

(back to top)