All case studies

AI Productivity · Agentic Systems

Worklode

Role-based AI operators that run real work inside one shared workspace.

The agent hub where teams run their agents from — voice-first operators that execute real database changes (creating, assigning, and rescheduling work across six views), each bounded by server-side validation so nothing runs the server hasn't checked.

BetaLive · the agent hub other projects run their agents from

Worklode

AI Productivity

By the numbers

23

Prisma models

111

Commits

6

Task views

Live

Tool-calling loop

The challenge

Traditional project tools force manual data entry for every change. Standard AI chatbots only produce text — they describe what to do but cannot alter your project board, so the human is still left doing the mechanical work.

The solution

An agentic tool-use execution loop that connects Anthropic Claude directly to a 23-model PostgreSQL database through typed server actions, so an operator can act on the board — bounded by server-side validation and live board context.

agent-work-os · tool-call.json
// Agentic Work OS · the model executes a real tool call
{
  "tool": "create_tasks",
  "input": {
    "project": "Launch",
    "tasks": [
      { "title": "Draft press release", "assignee": "sam" },
      { "title": "QA checkout flow", "priority": "high" }
    ]
  }
}
// → validated server-side, 2 tasks written, board refreshes

Illustrative: the model returns a typed tool call that the server validates and executes — real rows written to Postgres, board updated live.

Technical deep-dive

01

Architecture

Next.js 14 App Router, TypeScript, Prisma with a 23-model Supabase PostgreSQL database, Clerk auth with Svix webhooks, tRPC v11, and Zustand for client state.

02

AI agent loop

Claude function calling (create_tasks, update_tasks, schedule_huddle) with server-side execution and live board context injected into every turn.

03

Multimodal capture

Screenshot-to-task import using Claude Vision, with a human-in-the-loop verification step before anything is written to the board.

04

Voice integration

Native browser SpeechRecognition for dictation and ElevenLabs TTS for spoken operator responses — a genuinely voice-first workflow.

Product decisions

  • Operators mutate real records — not a chat that hands work back to the user.
  • Every tool call is typed and validated server-side before it touches the database.
  • Vision imports pass through human verification to prevent bad writes.
  • Voice-first as a design stance, not a bolt-on, informed by accessibility needs.

Built with

Next.js 14TypeScriptPrismaSupabase / PostgreSQLClerktRPC v11ZustandAnthropic ClaudeElevenLabs

Honest statusFeature-complete private build, pre-public launch.

Related guides

Want something like this?

Tell us what you’re building, the code you have, or the problem you need solved.

Start a project
Worklode — AI Productivity — Kruzeniski.ai · Kruzeniski.ai