Quick Start¶
Deploy the customer support template and call it over HTTPS. It takes about five minutes.
You need:
- an invite code (AgntSpark is invite-only during the alpha), and
- an API key for a model provider: OpenAI, Anthropic or Google.
1. Create your account¶
Open agntapi.agntspark.com, choose
Create account, and enter your name, email, a password and your invite
code. An invite link (…?invite=inv_…) fills the code in for you.
2. Create an agent from a template¶
- Go to Agents and click New Agent.
- Pick Customer Support.
- Give it a name, e.g.
support. - Keep the suggested model (
gpt-4o) or choose another, and paste that provider's API key into Model API key. - Leave Access on Private.
- Click Create & deploy.
The console shows the agent's access key (agk_…) and a ready-to-run
curl command. Copy the key now: it isn't shown again. You can always
create another one.
Your model key
The model key is stored encrypted and only given to this agent's containers. It is never shown again, in the console or the API.
The agent starts within a few seconds; its status turns running, and its
panel shows its URL, for example https://support-k3v9qa.run.agntspark.com.
3. Call it¶
export AGENT_KEY=agk_…
curl -X POST https://support-k3v9qa.run.agntspark.com/invoke \
-H "Authorization: Bearer $AGENT_KEY" \
-H 'content-type: application/json' \
-d '{"input": "I was charged twice this month."}'
{
"output": "I'm sorry about the double charge, that's frustrating. …",
"session_id": "5f0c…",
"agent_id": "agt_…"
}
To continue the conversation, send the session_id back:
curl -X POST https://support-k3v9qa.run.agntspark.com/invoke \
-H "Authorization: Bearer $AGENT_KEY" \
-H 'content-type: application/json' \
-d '{"input": "My email is sam@example.com", "session_id": "5f0c…"}'
Check that the agent is ready:
A 503 with "status": "unconfigured" means the agent has no working model
key. A plain-text 401 means the access key is missing or wrong. See
Runtime Contract v1.
4. Keys and access¶
In the agent's panel, under Access, you can create more keys (one per app, so you can revoke them separately), revoke keys, limit requests per caller, or make the agent public. See Access Keys and Rate Limits.