Releases: hcompai/hai-agents-python
Releases · hcompai/hai-agents-python
Release list
v1.0.2
v1.0.1
v1.0.0
v0.1.12
v0.1.11
v0.1.10
v0.1.8
Typed structured output
Pass a Pydantic model via answer_schema=; the SDK sends the matching JSON Schema as the agent's answer format and parses the final answer back into your model.
from pydantic import BaseModel
from hai_agents import Client, run_session
class Job(BaseModel):
title: str
company: str
class Jobs(BaseModel):
jobs: list[Job]
client = Client()
result = run_session(
client,
agent="h/web-surfer-holo3-1-35b",
messages="Find 3 open ML engineering roles in Paris.",
answer_schema=Jobs,
)
for job in result.answer.jobs: # result.answer is a Jobs instance
print(job.title, "@", job.company)Improvements
run_session/wait_for_sessionnow return when a session goes idle after answering, instead of polling until timeout.- Fixed environment payload serialization that could reject requests with a discriminator validation error.
v0.1.7
Custom tools
Pass local Python functions to an agent via tools=; the SDK executes the agent's tool calls on your machine and posts results back automatically.
def get_weather(city: str) -> str:
"""Current weather for a city."""
...
result = run_session(client, agent="h/web-surfer-holo3-1-35b", messages="...", tools=[get_weather])Also new: Session.agent_view_url, a link to the session's Agent View (live view and replay), and a clickable live-view link in the hai CLI.
Note: the v0.1.6 build on PyPI does not contain these features; this release is the one that ships them.