asiflow-hyper

Python client for the Hyper Agent API: delegate a software objective to Hyper, run and verify the work, read the evidence. Thin and honest: every method is one documented REST call (https://api.hyper.asiflow.ai/openapi.json); errors are the API's RFC 9457 problems.

from asiflow_hyper import HyperClient, device_login

token = device_login(client_name="my-agent", scope="objectives:read objectives:write runs:write evidence:read deploy:verify offline_access")
hyper = HyperClient(token=token.access_token)

job = hyper.create_objective(name="Bakery site", objective="A one-page site for Nour's bakery with a contact form.")
job = hyper.wait_job(job["id"])
run = hyper.start_run(job["result"]["project_id"], policy={"budget_cap_usd": 10})
run = hyper.wait_run(run["result"]["run_id"], on_needs_you=lambda card: "approved")
for attempt in run["attempts"]:
    if attempt["status"] == "verified":
        print(hyper.attempt_evidence(attempt["id"]))

Drive a session (1.1.0)

The interactive loop — the same one the Hyper web app runs — and every other capability:

s = hyper.sessions.create(repo="acme/site", permission_mode="auto_approve", budget_usd=10)
hyper.sessions.send(s["id"], "Add a pricing page, run the tests, open a pull request.")
view = hyper.sessions.wait_until_settled(s["id"])          # ready | waiting_input | …
if view["pending"] and view["pending"]["kind"] == "question":
    hyper.sessions.answer(s["id"], "blue")
    view = hyper.sessions.wait_until_settled(s["id"])
print(hyper.sessions.transcript(s["id"])["rows"][-1]["content"])
hyper.actions.invoke("publish_site", sessionId=s["id"])    # any of hyper.capabilities()["groups"]
hyper.sessions.end(s["id"])

Scopes sessions:read sessions:write (plus workspace:*, publish:write for the catalog). A person can hand you a token from Settings → Agent access on hyper.asiflow.ai, or your client registers itself and the person allows it on Hyper's consent page.

The agent-interop-e2e lane runs this client against production on every Hyper deploy.