Claude Code agents,
orchestrated.

Orchestrate multiple Claude Code sessions from one place. Give them memory, knowledge, and tools — then let them automate your projects while you focus on what matters.

Uses your existing Claude Code account. No extra API costs.

$ curl -fsSL https://raw.githubusercontent.com/gkkirsch/superbot3/main/install.sh | bash

How it works

localhost:3100

Spaces

api 6 tasks

landing 3 tasks

research idle

api space

Knowledge Workers

Build the auth system and deploy the landing page.

Reading knowledge... spawning 2 workers.

Read conventions.md. Using Zod + Express. 4/6 tasks done.

Landing deployed. API at 4/6 tasks complete.

Message the orchestrator...

auth-worker

$ claude --resume a3f2

Reading conventions.md

Scaffolding auth module

Implementing JWT

Writing refresh tokens

4/6 tasks done

deploy-worker

$ vercel deploy --prod

Building project...

Uploading assets

Ready app.vercel.app

db-worker

$ prisma migrate dev

Generating schema

Applied 3 migrations

Seeding test data

Database up to date

test-worker

$ npm test

42 tests passed

Coverage 94%

The problem

Every session starts from zero.

You explain your project structure. Again. You re-describe your conventions. Again. Your AI assistant has no memory, no context, and no idea what happened yesterday. You're the bottleneck in a system that's supposed to save you time.

The fix

Agents that accumulate intelligence.

Each space has persistent knowledge files, decision logs, and conventions. Workers read before they act. They write down what they learn. The next worker inherits everything. Day 30 is radically better than day 1.

The problem

Too many plugins to choose from.

MCP servers, plugins, skills — the ecosystem is exploding. Hundreds of options, half of them broken, and no way to know which ones actually work with Claude Code. You spend more time researching and configuring tools than using them. Every new space means starting the hunt from scratch.

The fix

Sensible defaults, fully customizable.

Superbot3 ships with curated, pre-configured plugins that just work — browser agent, design tools, web search, and more. Every space inherits sensible defaults out of the box. Then customize, swap, or add your own whenever you need to.

The problem

No visibility into what agents are doing.

Claude Code runs in a terminal. If you have multiple agents working, you're switching between tabs, scrolling through logs, and guessing what's happening. There's no overview, no status at a glance, and no way to drill into a specific agent without interrupting it.

The fix

A dashboard for every layer.

See all your spaces from the dispatcher, or drill down into an individual space orchestrator to watch its workers in real time. Conversation history, task progress, and agent status — all visible from a single dashboard without interrupting anything.

The problem

You can't easily control the system prompt.

Claude Code's CLAUDE.md gives you a starting point, but it's per-repo and per-session. You can't define a persistent personality, enforce a methodology, or give an agent a role that survives across sessions. Every worker starts with the same generic behavior.

The fix

Per-space CLAUDE.md you fully control.

Each space has its own CLAUDE.md that defines the orchestrator's personality, rules, and methodology. Workers inherit it automatically via CLAUDE_CONFIG_DIR. You control the system prompt at every layer — space-level instructions compose with repo-level conventions.

The problem

Scheduled tasks expire after 3 days.

Claude Code's built-in cron scheduler auto-expires recurring tasks after 7 days. Cloud-based triggers run on Anthropic's servers. You can't run agents locally on your own machine on a schedule that actually persists. Your crons just... stop.

The fix

Permanent crons, running locally.

Every scheduled task in superbot3 is permanent: true by default — they never auto-expire. Crons run locally on your machine inside your Claude Code session, not in the cloud. Daily reports, nightly backups, social media posts — they keep running as long as the space is alive.

The problem

You can't run agents while you sleep.

Claude Code is a conversation. You type, it responds. Close your laptop and everything stops. No way to schedule tasks, no background processing, no agents that run on their own. You're the human in the loop for everything, all the time.

The fix

Cron scheduling and full autonomy.

Each space has its own cron scheduler. Daily reports, nightly backups, social media posts at peak hours, email outreach on a cadence. Set the autonomy level per space — from "ask me everything" to "handle it all." Agents run 24/7 whether you're watching or not.

Architecture

How it works

Spaces communicate using Claude Code's built-in messaging inboxes. The dispatcher launches spaces, routes messages, and keeps everything alive.

Tier 1 — Dispatcher

Routes requests to the correct space

Dispatcher

Dashboard

file-based inboxes

Tier 2 — Space Orchestrators

Context, knowledge, and memory — uses Claude Code's agent teams to spawn workers

Facebook Campaign running

SB Landing Page running

GTM Research idle

tmux panes · ephemeral

Tier 3 — Workers

Gets the work done without polluting the orchestrator's context

auth-worker

db-worker

deploy-worker

test-worker

+ on demand

Use cases

One command. Entire workflows.

Build a full-stack app

API, database, auth, and frontend. Workers split the work, share conventions, and ship a PR.

Run social media

Draft posts, reply to comments, schedule content. A space that handles your entire social presence.

Monitor & respond

Scheduled agents that check systems, track competitors, or scan for issues — and act on what they find.

Deploy & maintain

CI/CD, infrastructure, and ongoing maintenance. Workers remember your deploy setup and repeat it perfectly.

Research & write

Market research, blog posts, email campaigns, documentation. Agents that write with context, not templates.

Run on a schedule

Cron-driven agents that execute tasks on your schedule. Daily reports, nightly backups, weekly digests.

Claude Code agents,
orchestrated.

Every session starts from zero.

Agents that accumulate intelligence.

Too many plugins to choose from.

Sensible defaults, fully customizable.

No visibility into what agents are doing.

A dashboard for every layer.

You can't easily control the system prompt.

Per-space CLAUDE.md you fully control.

Scheduled tasks expire after 3 days.

Permanent crons, running locally.

You can't run agents while you sleep.

Cron scheduling and full autonomy.

How it works

One command. Entire workflows.

Build a full-stack app

Run social media

Monitor & respond

Deploy & maintain

Research & write

Run on a schedule

Claude Code is powerful.
Superbot3 makes it persistent.

Get started in 60 seconds

Stop re-explaining.
Start compounding.

Claude Code agents,orchestrated.

Every session starts from zero.

Agents that accumulate intelligence.

Too many plugins to choose from.

Sensible defaults, fully customizable.

No visibility into what agents are doing.

A dashboard for every layer.

You can't easily control the system prompt.

Per-space CLAUDE.md you fully control.

Scheduled tasks expire after 3 days.

Permanent crons, running locally.

You can't run agents while you sleep.

Cron scheduling and full autonomy.

How it works

One command. Entire workflows.

Build a full-stack app

Run social media

Monitor & respond

Deploy & maintain

Research & write

Run on a schedule

Claude Code is powerful.Superbot3 makes it persistent.

Get started in 60 seconds

Stop re-explaining.Start compounding.

Claude Code agents,
orchestrated.

Claude Code is powerful.
Superbot3 makes it persistent.

Stop re-explaining.
Start compounding.