Introduction

Autograd begins as an inference platform for agents and grows into a reinforcement learning loop built from their real work.

Why start with inference?

Agent teams need a capable model before they need a training system. Routing inference through Autograd creates immediate value and establishes a consistent record of model calls, tool use, latency, token usage, and model versions.

The SDK adds the signals that are only visible inside your product: whether a task completed, whether a user accepted or corrected the result, and which reward or grader outcome belongs to the run.

Quickstart

The planned inference interface follows OpenAI client semantics, allowing an existing agent to start with a base URL and model change.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.autograd.run/v1",
    api_key=os.environ["AUTOGRAD_API_KEY"],
)

response = client.chat.completions.create(
    model="qwen-3.8-27b",
    messages=[
        {"role": "user", "content": "Plan the task."}
    ],
)

print(response.choices[0].message.content)

Authentication

The intended client reads an Autograd API key from the environment. Key creation, rotation, and organization-level access controls will be documented when preview access opens.

Qwen 3.8 27B

Qwen 3.8 27B is the initial model planned for agent inference. Context length, output speed, supported tool-call behavior, concurrency, regions, and pricing will be published from measured production configurations—not estimates.

Chat completions

The first endpoint is designed around the familiar chat-completions request shape. The final request schema, streaming format, usage fields, and error envelope will be documented from the deployed API.

Traces and rewards

Inference captures the execution. Your application supplies the meaning. The proposed SDK links product events and reward signals to the same run.

with autograd.trace() as run:
    result = agent.invoke(task)
    run.reward("task_success", 1.0)
    run.event("user.accepted")

SDK names and signatures shown here are directional and not yet a compatibility promise.

How the reinforcement loop works

1. RunCapture structured agent executions through inference.
2. RewardJoin outcomes, corrections, graders, and training permissions.
3. ReinforceEvaluate a baseline, train on selected work, and evaluate the candidate.
4. PromoteApprove a better checkpoint behind the same inference interface.