Quickstart

Point an OpenAI client at Zumik

The only line that changes is the base URL. Request and response bodies stay byte-for-byte OpenAI-compatible.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.zumik.ai/v1",
    api_key="zk_live_...",
)

r = client.responses.create(
    model="code.balanced",   # a Zumik alias, resolved at request time
    input="Review the latest patch.",
)
print(r.output_text)

Create a reusable artifact on /v2

Turn stable instructions, tool definitions, or policies into an opaque, immutable handle instead of resending them on every call. The handle is the artifact ID; the content lives behind the tenant boundary.

curl https://api.zumik.ai/v2/artifacts \
  -H "Authorization: Bearer zk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"artifact_type":"policy","content":"Run the linter before every commit."}'
# => { "id": "art_01JY...", "object": "artifact", "artifact_type": "policy" }

Group artifacts into an ordered bundle, then attach the bundle to a session to carry stable state across turns.

Run a workload diagnostic

Before you change any infrastructure, score how much of your traffic is reusable. The diagnostic reads metadata traces and returns a Workload Reuse Score, the reuse waterfall, a recommended execution profile, and the missed-opportunity gap.

curl https://api.zumik.ai/v2/diagnostics \
  -H "Authorization: Bearer zk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"source":"trace_export","trace_mode":"metadata","sample_ref":"trc_..."}'

The diagnostic runs on metadata only by default, so no raw prompt text is retained. See workload diagnostics for how to read the report.