Register your first agent

Assumes you’ve already finished the install and the dashboard is up at http://localhost:3000.

1. Sign in with the bootstrap admin credentials the installer printed. They also live in <install-dir>/.env:

   grep -E '^AGENTUM_ADMIN_(EMAIL|PASSWORD)=' .env

2. Open the Agents page in the dashboard (/agents) and click Register agent. Fill in:

   • Name — friendly identifier, e.g. support-bot.
   • Owner email — the human responsible (alice@yourco.com).
   • Purpose — one-line description, surfaces in audit and in HITL prompts.
   • Declared tools — comma-separated list of tools the agent will call (search_kb, create_ticket). Optional but recommended: it lets you write policies referencing tool names.

   Or use the CLI:

   docker exec lupid-agentum-1 agentum agent register \
     --name support-bot \
     --owner-email alice@yourco.com \
     --purpose "answers customer questions, opens Zendesk tickets"

3. Read the response. You’ll get back:

   {
     "agent_id": "9da9afd6-7158-42b1-ada4-9c84dd4c4d67",
     "name": "support-bot",
     "status": "active",
     "session_jwt": "eyJ0eXAiOiJKV1QiLCJhbGciOiJFZERTQSJ9..."
   }

   The session_jwt is the agent’s bootstrap session token. Save it.

4. Wire it up. Pick one of the two paths:

   SDK (TypeScript)

   import { Lupid } from "@lupid/agentum-sdk";
   
   const lupid = await Lupid.init({
     baseUrl:   "http://localhost:3000",        // dashboard URL — proxies to API
     apiKey:    process.env.LUPID_API_KEY!,     // mint via `agentum admin api-keys mint`
     agentName: "support-bot",
     declaredTools: ["search_kb", "create_ticket"],
   });
   
   // Every fetch to a recognised LLM host (OpenAI, Anthropic, Cohere, Gemini,
   // Azure OpenAI, and the OpenAI-compatible providers Together / Mistral /
   // Groq / DeepSeek / X.AI / Perplexity / OpenRouter / Fireworks / Anyscale)
   // now routes through Lupid.
   const completion = await fetch("https://api.openai.com/v1/chat/completions", {
     method: "POST",
     headers: { Authorization: `Bearer ${process.env.OPENAI_API_KEY}` },
     body:    JSON.stringify({ model: "gpt-4o", messages: [...] }),
   });

   Network proxy

   In whatever environment runs the agent (Docker compose, K8s pod, plain process):

   export HTTPS_PROXY=http://lupid:7070
   export AGENTUM_AGENT_ID=9da9afd6-7158-42b1-ada4-9c84dd4c4d67
   export AGENTUM_SESSION_JWT="eyJ0eXAi..."

   No code changes. The agent’s outbound HTTPS calls issue a CONNECT to Lupid, which intercepts the TLS handshake and enforces the policy.

5. Watch it. Open the dashboard’s audit page (/audit). The first request your agent makes lands as a tool_call row within a second.

What’s next

Now that the agent is registered, the next decisions are:

• Write a policy that fits the agent’s risk profile — by default a fresh agent in enforce mode has no policy and every call is denied. See Writing a policy.
• Store any credentials the agent uses (Stripe key, database password) in the vault so the agent doesn’t carry long-lived secrets. See Credential vault.
• Configure HITL for the high-risk subset of tool calls. See the security concepts page.