No Code — AgentBreeder Studio

Build, configure, and deploy production agents entirely from AgentBreeder Studio (formerly the Dashboard). No YAML. No code. No DevOps knowledge required.

Who this is for: Product managers, analysts, business users, executives, and anyone who wants to ship agents without touching a terminal.

Supported Frameworks

Every framework works in Studio. Pick the one that fits your use case from a dropdown — AgentBreeder handles the runtime scaffolding automatically.

Supported Deployment Targets

All deployment targets are available from the Deploy to dropdown. Connect your cloud account once in Settings — then deploy anywhere without manual credential configuration.

Status: Deploy-target availability (which targets ship full governance, greenfield provisioning, and secret auto-mirror) is maintained in one canonical place — see the deploy-target status table.

Getting Started

1. Open Studio

http://localhost:3001        # local dev stack

Or log in to your org's hosted AgentBreeder instance.

Default local credentials:

2. Create a new agent

Click New Agent in the top-right corner. The visual builder opens.

3. Configure in the form

The builder asks you for:

4. Deploy

Click Deploy Agent. Studio runs the 8-step pipeline and shows live progress:

✅  YAML parsed & validated
✅  RBAC check passed
✅  Dependencies resolved
✅  Container built
✅  Deployed
✅  Health check passed
✅  Registered in org registry
✅  Endpoint returned

If any step fails, the deploy rolls back automatically.

5. Your agent is live

The endpoint URL appears in Studio. You can immediately:

• Chat with the agent from the playground tab
• View logs in real time
• Monitor costs per token, per day, per team
• Share the endpoint URL with your team

MCP Servers

Studio has a built-in MCP server registry. From the agent builder:

1. Click Add Tool → Search MCP Servers
2. Browse your org's registered MCP servers or auto-discover from your connected cloud
3. Click Add — the server is wired in as a sidecar automatically on deploy

You can also run agentbreeder scan from the CLI to discover locally running MCP servers and publish them to the registry for Studio to find.

Register an agent from YAML

If you already have an agent.yaml (handed over from an engineer, or pasted from another project), you can register it without ever leaving Studio.

1. Go to http://localhost:3001/agents
2. Click Register from YAML — the button sits next to Create Agent at the top right
3. The page at http://localhost:3001/agents/register opens with a YAML editor
4. Paste your agent.yaml, click Validate (parses + JSON Schema check), then Submit

The agent record is persisted via POST /api/v1/agents/from-yaml and stores the name, version, framework, deploy config, and refs to prompts + tools. Once registered, the agent appears in the registry and can be deployed from Studio like any other agent.

Cross-references inside the YAML resolve from the same registry Studio uses:

• Prompts: prompts: {system: prompts/microlearning-system} resolves to the prompt managed at Prompts →
• Tools: tools: [{ref: tools/web-search}, {ref: tools/markdown-writer}] resolves to entries in the tools registry

Try a prompt without code

The Test tab on a prompt page now uses a real Gemini call (it was simulated previously).

1. Go to http://localhost:3001/prompts/<id> — for example, the microlearning-system prompt
2. Click the Test tab
3. Enter a user message in the chat box (e.g. "What is your job?")
4. Click Send Test

Behind the scenes Studio calls POST /api/v1/registry/prompts/{id}/render with {user_message, model, temperature} and renders the live LLM response. The default model is gemini-2.5-flash and the response includes the model used and total duration in milliseconds.

Run a tool from Studio

Every registered tool gets an auto-generated form and runner UI.

1. Go to http://localhost:3001/tools/<id> — for example, the web-search or markdown-writer tool
2. Click the Try it tab
3. The form is auto-generated from the tool's JSON schema — fill the fields
4. Click Run tool

The result panel shows:

• output — the structured tool result
• stdout / stderr — raw process streams (for python: and node: dispatch)
• exit_code — process exit code (0 on success)
• duration_ms — wall-clock duration (e.g. ~25 ms for in-process Python, ~1.5 s for a cold Node tool)

Calls go through POST /api/v1/registry/tools/{tool_id}/execute, which dispatches by the tool's endpoint field (engine.tools.standard.<name>, python:<abs_path>, node:<abs_path>, or HTTP).

Chat with a deployed agent

Once an agent is deployed (locally or to Cloud Run / ECS / etc.), Studio talks to it through a server-side proxy — no CORS pain, and bearer tokens stay server-side.

1. Go to http://localhost:3001/agents/<id> — for example, microlearning-ebook-agent
2. Click the Invoke tab
3. Fill in:
   • Endpoint — e.g. https://microlearning-ebook-agent-sizukgalta-uc.a.run.app
   • Bearer token — pasted directly, or pulled from AGENT_<UPPER_SNAKE>_TOKEN env var server-side
   • Message — the user input (e.g. "What is your job?")
4. Click Send

You get back the agent's output, the session_id (auto-filled into the form for the next turn so multi-turn chat just works), and the duration_ms and status_code from the upstream call. The proxy is POST /api/v1/agents/{agent_id}/invoke.

Multi-Agent Orchestration (Visual Canvas)

Studio includes a ReactFlow-based drag-and-drop canvas for building multi-agent pipelines — no code required.

To open the canvas:

1. Click New Orchestration in the top-right corner
2. Drag agents from the registry onto the canvas
3. Draw connections between them to define the flow
4. Set the orchestration strategy (router, sequential, parallel, supervisor)
5. Click Deploy Orchestration

The canvas generates an orchestration.yaml behind the scenes. At any point, click View YAML to see it.

Human-in-the-Loop Approvals

For sensitive workflows, you can require a human to approve before an agent proceeds:

1. Open the agent builder → Advanced → Approvals
2. Enable Require HITL approval
3. Set the approvers (team or individuals)

When the agent hits an approval step, it pauses and sends a notification. The approver sees the pending action in Studio and can approve or reject.

Editing an Agent

Open an agent from the registry → click Edit → change any field → Redeploy.

Every redeploy is versioned. You can roll back to any previous version from the History tab.

Eject to Low Code (View the YAML)

At any point, click View YAML to see the agent.yaml that Studio generated. This is human-readable and version-controllable.

To take full ownership of the YAML:

1. Click Export agent.yaml
2. Run agentbreeder deploy agent.yaml from your terminal
3. You are now in Low Code mode — see Low Code →

The exported YAML is identical to what Studio used. Nothing is hidden.

Tier Mobility

No Code (Studio)
  └─▶  View YAML            → copy into your IDE    → Low Code
  └─▶  agentbreeder eject   → full SDK scaffold      → Full Code

You can always go deeper — you never lose what you built. See Full Code →.

All Frameworks × All Deployment Targets

Every framework compiles to every deployment target. The full matrix — and the per-target governance caveats — live in the agent.yaml reference → All Frameworks × All Deployment Targets.

Next Steps