Appearance

5-Minute Agent Endpoint Quickstart ​

Create and call a managed Hermes Endpoint without creating an Agent, Environment, or Sandbox first. This example requires curl and jq.

Create and invoke an Endpoint ​

bash
export SANDBASE_API_KEY='sk-sb-YOUR_KEY'
export SANDBASE_API_BASE='https://api.sandbase.ai'

: "${SANDBASE_API_KEY:?SANDBASE_API_KEY is required}"

ENDPOINT_RESPONSE="$(
  curl -fsS -X POST "$SANDBASE_API_BASE/v1/endpoints" \
    -H "Authorization: Bearer $SANDBASE_API_KEY" \
    -H 'Content-Type: application/json' \
    --data '{"name":"quickstart-assistant","runtime":"hermes"}'
)"

ENDPOINT_ID="$(jq -er '.id' <<<"$ENDPOINT_RESPONSE")"
RUN_URL="$(jq -er '.run_url' <<<"$ENDPOINT_RESPONSE")"

jq '{id, runtime, status, run_url, mcp_url}' <<<"$ENDPOINT_RESPONSE"

curl -fsS --max-time 240 -X POST "$RUN_URL" \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H 'Content-Type: application/json' \
  -H 'Prefer: wait' \
  --data '{
    "input": "Remember that my project is Atlas, then confirm what you remembered.",
    "thread_id": "quickstart-customer-123"
  }' \
  | jq .

The first call may take longer than a warm call while SandBase starts the Sandbox and prepares the runtime. Reuse the same thread_id to continue the durable conversation; omit it for a stateless, single-turn call.

Add Skills and MCP tools ​

To add your own capabilities, create and upload a Skill and put the returned .name value in the Endpoint skills list. Configure remote HTTPS tools with mcp_servers; see Create & Invoke Endpoints for the complete declarative YAML definition and protocol reference.

Clean up ​

bash
curl -fsS -X DELETE "$SANDBASE_API_BASE/v1/endpoints/$ENDPOINT_ID" \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  | jq .

Next steps ​