Documentation

Getting started, CLI reference, and architecture overview.

Quick Start

# 1. Initialize project
orgsdk init

# 2. Create an agent
orgsdk agent create my-agent \
  --role '{"name":"my-agent","description":"Does X"}' \
  --plugins core data

# 3. Run a one-shot task
orgsdk agent --agent my-agent --one-shot "Summarize docs"

# 4. Start the daemon
orgsdk daemon start

# 5. Launch the dashboard
orgsdk dashboard

CLI Reference

orgsdk init [dir] [--url <postgres-url>]

Initialize project and connect to Postgres

orgsdk agent create <name> --role '<json>' [--plugins id1 id2]

Create a new agent with role and plugins

orgsdk agent --agent <name> [prompt...]

Interactive session with an agent

orgsdk workflow run <agent> <workflow> [--params '{...}']

Run a workflow on an agent

orgsdk schedule add <name> --agent-id <id> --workflow-id <id> --cron '...'

Add a cron schedule

orgsdk daemon start [--detach]

Start the background daemon

orgsdk dashboard [--port 3839]

Start the web dashboard

orgsdk approve watch [--agent <name>]

Watch for and handle pending approvals

Architecture

OrgSDK is built around a simple principle: all state lives in Postgres. Agents, workflows, schedules, events, approvals, credentials, artifacts, sessions, and memory are all stored in relational tables.

Workflows execute in an isolated VM with access to the agent's plugin tools. The VM is sandboxed — no filesystem or network access outside of the plugin APIs. This means workflows can safely call LLMs, browse the web, read files, and emit events without breaking the host.

The daemon polls for schedules and events, executing them when due. The dashboard serves a real-time WebSocket + HTTP interface over the same Postgres connection.