Put an AI agent on the internet. Keep it private.

# 1. Deploy a template
curl -X POST https://agntapi.agntspark.com/v1/agents \
  -H "Authorization: Bearer $AGNTSPARK_API_KEY" \
  -d '{"name": "support", "model": "gpt-4o",
       "api_key": "sk-…",
       "deploy": {"image": "agntspark/template-customer-support"}}'

# → {"url": "https://support-k3v9qa.run.agntspark.com",
#    "access": "private", "access_key": "agk_…"}

# 2. Call it with its key
curl -X POST https://support-k3v9qa.run.agntspark.com/invoke \
  -H "Authorization: Bearer agk_…" \
  -d '{"input": "I was charged twice this month."}'

# → {"output": "Sorry about that — I've opened ticket
#    T-1042 with billing…", "session_id": "5f0c…"}

from agntspark import Client, DeployConfig

with Client() as client:
    agent = client.agents.create(
        "support",
        model="gpt-4o",
        api_key=OPENAI_API_KEY,
        deploy=DeployConfig(
            image="agntspark/template-customer-support",
        ),
    )
    # Private by default: the first key is returned once.
    key = agent.access_key

    reply = client.agents.invoke(
        agent, "I was charged twice this month.",
        access_key=key,
    )
    print(reply.output)

$ agntspark deploy agent.yaml --wait
✓ Agent created: agt_7Hq2…
  URL      https://support-k3v9qa.run.agntspark.com
  Access   private
  Access key (shown once): agk_…

$ agntspark invoke agt_7Hq2… "Where is my refund?" --key agk_…
Refunds go back to the original card within 5–10…

$ agntspark access agt_7Hq2… private --rpm 30
agt_7Hq2…: private, 30 requests/min per caller

$ agntspark keys create agt_7Hq2… --label website