ASC Cloud

Building agents on ASC

Versioned prompts, tool registry, A/B routing, eval harness, and unified trace — without gluing five vendors together.

The ASC agent runtime gives you everything you'd otherwise build on top of LangChain + Langfuse + a feature flag service + a vector store

a queue. It's all there, with one auth, one billing meter, and one trace.

Define an agent

from asc import Client
 
asc = Client(api_key="asc_...")
 
agent = asc.agents.create(
    name="support-bot",
    system_prompt="You are a polite, concise support assistant.",
    tools=["lookup_order", "refund"],
    model="claude-sonnet-4",
)

Run it

run = asc.agents.run(agent["id"], input="where is order #1234?")
print(run["output"])

Versioned prompts + A/B routing

v2 = asc.agents.create_version(
    agent["id"],
    system_prompt="You are an empathetic support assistant.",
    weight=0.2,  # 20% of traffic
)

20 % of new runs route to v2, 80 % stay on the prior version. Deltas show up in asc.evals and the trace dashboard.

Eval suites

suite = asc.evals.create_suite("support-golden", scorer="contains")
asc.evals.add_case(suite["id"], input="where is order #1234?", expected="tracking")
asc.evals.add_case(suite["id"], input="refund my order", expected="refund")
 
run = asc.evals.run(suite["id"], agent_id=agent["id"])
print(run["score"])  # 0.0 — 1.0

Trace

asc.trace.get(run["agent_runs"][0]["id"])
# returns the full step graph: prompt → tool calls → token usage

← ASC Cloud — Overview Governance: firewall · residency · audit · teams →