Skip to content

cdsaidev/Agent-Foundry

Repository files navigation

   ╭────────────────────────────────────────────╮
   │              Agent Foundry                 │
   │   Task-native assistants you ship & share  │
   ╰────────────────────────────────────────────╯

Agent Foundry

One line: Build focused automations with Claude, list them in a store, install what others publish, and review incoming submissions—with streaming generation along the way.

ss


Contents

  1. Why this exists
  2. What you can do today
  3. What is still rough
  4. Stack assumptions
  5. First run
  6. Scripts you will actually use
  7. Configuration surface
  8. License

Why this exists

Most “assistant” products optimize for open-ended chat. Agent Foundry optimizes for narrow jobs: you describe a task, the model drafts a runnable package of behavior, and the app treats that artifact as something you can version, publish, and reuse—not a one-off reply.


What you can do today

Area Status
Generate task-specific assistants from prompts Done
Browse a shared catalog (“store”) Done
Publish your own builds Done
Install items from the catalog locally Done
Moderation / review queue for submissions Done
Token streaming while generating Done

What is still rough

  • Sandboxed execution for arbitrary scripts (planned)
  • “Advanced” authoring studio (planned)
  • Multi-assistant chat sessions (planned)
  • Catalog pagination (acknowledged gap)

Stack assumptions

  • Runtime / package manager: Bun is the happy path.
  • Database: PostgreSQL, accessed through Prisma migrations.

If you prefer another JS runtime, you are on your own—commands below assume bun.


First run

Expand: clone → env → database
  1. Clone the repository you are reading from.

  2. Install dependencies:

    bun install
  3. Copy the example environment file and edit values:

    cp .env.example .env
  4. Create an empty PostgreSQL database and point DATABASE_URL at it.

  5. Apply schema and regenerate the Prisma client:

    bun prisma migrate deploy && bun prisma generate

Scripts you will actually use

Goal Command
Local dev server bun run dev
Production bundle + serve bun run build && bun run start

Configuration surface

These keys are expected in .env. Names are literal; descriptions are purpose-focused.

Key Role
NODE_ENV development vs production behavior switches
DATABASE_URL PostgreSQL connection string for Prisma
SUPERADMINS Comma-separated emails with elevated privileges
SECRET Application secret for signing / crypto
GOOGLE_CLIENT_ID OAuth client id
GOOGLE_CLIENT_SECRET OAuth client secret
GOOGLE_REDIRECT_URI OAuth redirect registered in Google Cloud
GOOGLE_SEARCH_KEY Programmatic web search credential
GOOGLE_SEARCH_CX Programmatic search engine id (cx)
BROWSER_URL Base URL for the headless browser microservice
BROWSER_SECRET Shared secret for that microservice
SCRIPT_URL Base URL for the remote script runner
SCRIPT_SECRET Shared secret for the script runner
ANTHROPIC_API_KEY Claude API access

Tip: Never commit a populated .env. The committed .env.example is the contract; your local file is private state.


License

Distributed under the terms in LICENSE.md (MIT).

About

Build, publish, and install task-focused AI agents with Claude. No-code studio, shared agent store, streaming generation, and admin moderation. Next.js + PostgreSQL.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages