⬡ Concentus GitHub
Visual multi-agent orchestration

Wire up a team of AI agents.
Run them on your own Claude.

A visual canvas for building multi-agent flows. Design them, connect your tools and databases, and run on your Claude subscription — or the cloud. No API key needed to start.

Get started View on GitHub
Runs on your Claude login MCP + databases Local or cloud Steer live
Powered by Claude/ Opus 4.8 & Sonnet 5/ the Claude CLI/ Anthropic Managed Agents
What's inside

Everything an agent pipeline needs — on one canvas.

Draw the flow, wire in your tools and data, and run it. No glue code, no orchestration boilerplate.

Canvas

Canvas, not config

Drop agents, tools and data sources onto a canvas and connect them. The flow you draw is exactly the flow that runs — nothing hidden in YAML.

Orchestration

A coordinator and a crew

One coordinator plans the work and delegates to specialized sub-agents that run in parallel — then reconciles their results.

Your Claude

Run on your subscription

Executes locally through the Claude CLI on your Pro or Max plan. Add an API key only when you want the cloud.

Tools

Bring your world in

Connect MCP servers like GitHub and Linear. Studio registers and authorizes them for you.

Data

Answers from your database

Point a SQL node at any JDBC database; its rows are retrieved and injected into agent context at run time.

Library

Reusable resources

Save agents, MCP servers, and database connections once — then select them in any flow.

Control

Steer it live

Watch output stream in and send commands to a running flow — without stopping it.

Efficiency

Lean by default

Cheaper models for volume, stronger ones for judgment, and prompts written to spend tokens carefully.

Triggers

Start it from anywhere

Run by hand, on a cron schedule, or from a webhook — a Linear issue or a GitHub push kicks off a flow with the event as its input. Deliveries are verified by HMAC signature against the provider's own secret, so any provider works without provider-specific setup.

Editing

Copy, paste, duplicate

Build a pattern once and reuse it. Copy blocks with the connections between them intact, paste them into another flow, or duplicate a whole flow from its card. Copies come back wired up and ready to edit — never sharing an identity with the original.

Run anywhere

Your subscription today. The cloud when you need it.

The same flow runs both ways — flip the switch with a single environment variable.

Local · default

On your Claude subscription

Runs through the claude CLI you already use. Your prompts, code, and data stay on your machine — only the model calls go to Anthropic.

  • No API key, no per-token bill
  • Sub-agents map to Claude Code subagents
  • MCP servers auto-registered & authorized
Cloud · optional

In a managed sandbox

Set an API key and the same flow runs as an Anthropic Managed Agents session — a hosted sandbox that clones repos, runs tools, and scales past your laptop.

  • One coordinator + a versioned sub-agent roster
  • Repos mounted, tools sandboxed
  • Same canvas, no rewrite
How it works

From blank canvas to running crew in four moves.

1

Design

Drop agent, MCP, database, and repo nodes and connect them. Mark one agent the coordinator.

2

Connect

Pick tools and data from your resource library — or add a new one in a couple of fields.

3

Run

Launch on your subscription or in the cloud. Output streams into the console live.

4

Steer

Send commands mid-run, review what each agent did, and iterate without restarting.

Use cases

Wire the crew for the job.

linear→lead→befe

Ship from a ticket

Pull a story from Linear, split the backend and frontend work across sub-agents, and push the changes.

repo→reviewer

Audit a diff

A reviewer agent sweeps changed files for correctness bugs and security issues, ranked by severity.

postgres→agent

Data-aware answers

An agent queries your database and answers with the real numbers — not a guess from training data.

Install

Run it your way — laptop to cluster.

Open source. Clone the repo and pick a method. Every path runs on your Claude subscription — no API key required.

Local dev

Two terminals

The designer and API start with no credentials. Sign in to Claude once to run flows.

claude            # sign in once (subscription)
cd apps/backend && mvn spring-boot:run
pnpm install && pnpm dev   # http://localhost:5173
Docker

Compose up

Backend (with the Claude CLI baked in) + frontend. Mint a subscription token once, no API key.

claude setup-token        # prints a token
echo "CLAUDE_CODE_OAUTH_TOKEN=..." > .env
docker compose up --build # :3000
Kubernetes

Helm chart

One release for backend + frontend, with an optional public nginx entrypoint.

helm install concentus deploy/helm/concentus \
  --set backend.claudeOAuthToken=$TOKEN \
  --set publicNginx.enabled=true
Kubernetes

Kustomize

Plain manifests. The public overlay adds the optional nginx to expose it externally.

kubectl apply -k deploy/kustomize/base
# or, with the public nginx entrypoint:
kubectl apply -k deploy/kustomize/overlays/public

Full walkthrough, values, and the webhook setup are in the README.

Build your first flow in minutes.

Sign in to Claude once, open the canvas, and run a multi-agent flow on your own subscription — no API key, no setup tax.

View on GitHub Installation methods