Getting Started with Dobby AI in 5 Minutes

AI agents are transforming how teams build software. But as the number of agents grows, so does the complexity of managing them. Dobby AI was built to solve exactly this problem -- giving you a single control plane for all your AI agents, regardless of which framework they use. Whether you are running CrewAI crews, LangChain chains, OpenAI assistants, or custom Python agents, Dobby brings them all under one roof with governance, observability, and cost control built in.

This guide walks you through the five steps to go from zero to a fully operational agent workspace. The entire process takes about five minutes.

Step 1: Create Your Workspace

After signing up with your Google account, Dobby walks you through a guided onboarding wizard. The first decision is creating a workspace (called a "tenant" in the platform). Each workspace is fully isolated -- its own agents, tasks, policies, and audit trail. You will choose a name and, critically, your data residency region: Israel, EU, or US. This choice is permanent and determines where all your data is physically stored, which matters for GDPR and data sovereignty compliance.

During onboarding, you will also select your plan tier. Each tier comes with pre-configured governance templates -- Conservative, Balanced, Team, or Enterprise -- that set sensible defaults for agent autonomy, safety limits, and cost controls. The Conservative template requires human approval for every agent action. The Enterprise template gives agents broad autonomy with budget guardrails. You can customize all of these later, so do not overthink the initial choice.

Step 2: Make Your First LLM Call

The fastest way to see value is through the Agentic Gateway. Navigate to the Gateway tab in your dashboard and click Create Key. This generates a gateway API key that you can use with the standard OpenAI SDK -- no custom Dobby SDK required. Here is a complete example that makes an LLM call through Dobby:

from openai import OpenAI

# Point the standard OpenAI SDK at Dobby's gateway
client = OpenAI(
    api_key="gk_user_your_key_here",
    base_url="https://dobby-ai.com/api/v1/gateway"
)

response = client.chat.completions.create(
    model="claude-sonnet-4",  # or gpt-4o, gemini-2.5-flash
    messages=[{"role": "user", "content": "Summarize this PR diff"}]
)

print(response.choices[0].message.content)

Every request through the gateway is automatically authenticated, metered, and logged. You will see the cost, token usage, and latency appear in your dashboard within seconds. Switch between providers by changing the model parameter -- no code changes, no new SDK, no reconfiguration. This is the core value proposition: use the tools you already know, but with full visibility and control layered on top.

Step 3: Connect Your AI Provider

Dobby supports 13 or more LLM providers out of the box: Claude, GPT-4, Gemini, Mistral, Llama, DeepSeek, and more. Navigate to Connections and then AI Models, select your provider, and paste your API key. All credentials are encrypted with AES-256-GCM using tenant-derived encryption keys, so your secrets are protected both in transit and at rest. You can configure a default model that agents will use unless overridden, and set up fallback providers for resilience.

The best AI platform is the one that works with what you already have. No lock-in, no proprietary SDKs -- just bring your agents and your API keys.

Step 4: Set Up Governance

Before agents run at full speed, define your rules. Go to Settings and then Policies to configure three key areas. First, model restrictions -- specify which LLM models are approved for use and block preview or experimental models from production workloads. Second, cost limits -- set daily and monthly budgets per tenant and per agent, with automatic alerts at 80 percent, 90 percent, and 100 percent thresholds. Third, approval requirements -- define which actions require human sign-off before an agent can proceed. This ensures no agent exceeds your budget, uses unauthorized models, or takes high-risk actions without oversight.

Step 5: Create Your First Task

Navigate to Activity and click Create Task. Describe what you need done in plain language, set the priority level, and assign it. Dobby's task engine auto-assigns the best agent based on capability matching -- if you have multiple agents registered, it picks the one whose skills match the task description. Every action the agent takes is logged to an immutable audit trail, and you can track progress in real time via SSE streaming. If the agent hits an approval gate, it pauses and notifies you via the dashboard or Slack.

Step 6: Monitor Everything

As your agents work, the Insights dashboard gives you full visibility into costs, token usage, and agent performance. The FinOps page breaks down spend by provider, by agent, by user, and by department -- so finance teams can see exactly where AI dollars go. The Live view shows a real-time feed of every gateway request with model, tokens, cost, and latency. The Agent Fleet view shows the health and status of every registered agent. Set up Slack notifications to get alerted on budget thresholds, approval requests, and agent errors without needing to watch the dashboard constantly.

What Comes Next

Once your first task completes successfully, you have the foundation in place. From here, explore scheduling for recurring automation -- agents can run on cron schedules or be triggered by events. Try the MCP protocol integration to connect agents to external tools like GitHub, Jira, and Slack directly from Claude Desktop or Cursor. Set up the Agent Control Bar for at-a-glance operational intelligence on every dashboard page. Browse the Skill Marketplace to install pre-built agent capabilities like code review, security scanning, and QA testing. And invite your team members with role-based access control so everyone -- from developers to executives -- has the right level of visibility and control. The entire Dobby platform is designed to grow with your agent fleet, from your first experiment to your hundredth production agent.