Human approval API

Put a human decision inside any workflow.

One server-authoritative primitive for AI agents, scripts, CI/CD pipelines, cron jobs, and high-impact operations. Ask with typed context, wait safely, and resume with a signed result.

Create a free Source Read the 5-minute guide

Install the CLI

One signed binary for macOS, Linux, and Windows.

Configure a Source

Store the one-time Source key locally with owner-only permissions.

Ask and wait

Pause cleanly until a person returns a typed decision.

From shell to signed answer

A blocking call without a busy loop.

The CLI creates an Action, waits with bounded long-polling, and returns a machine-readable selection once a human decides.

Install

python -m pip install actionbox
actionbox configure
actionbox doctor

Ask from any script

actionbox ask "Deploy to production?" \
  --option approve=Approve \
  --option reject=Reject \
  --wait --timeout 30m

Native SDKs

Use the same protocol from application code.

Create typed human checkpoints without inventing polling, state, expiry, or callback verification.

Python

pip install actionbox-sdk

from actionbox import Actionbox

client = Actionbox()
decision = client.ask(
    "Deploy v2.18?",
    options=["approve", "reject"],
    wait=True,
)

AI agents

CI/CD

Background jobs

Risky operations

Design beyond the endpoint

Build the control system around the human decision.

The API provides the durable decision boundary. These guides cover the policy, execution state, and identity controls your integration still owns.

AI agent governance

Choose which operations run, stop, or require human judgment.

Read the guide

AI agent orchestration

Coordinate checkpoints, approvals, retries, and execution outcomes.

Read the guide

AI agent identity

Separate workload access, reviewer authority, and downstream execution.

Read the guide