# Actionbox

> Actionbox is a server-authoritative human decision layer for automation and AI agents.

## Overview
- [Documentation](https://actionbox.cloud/docs): Full developer guide, API reference, and SDK tutorials.
- [Human Approval API](https://actionbox.cloud/human-approval-api): Add a durable approval gate to AI agents, scripts, and automations.
- [Slack approvals](https://actionbox.cloud/slack): Review requests and follow results in Slack; assisted Team beta setup required.
- [ActionBox for iPhone](https://actionbox.cloud/mobile): Review Actions and configure optional push alerts from the iPhone app.
- [Pricing](https://actionbox.cloud/pricing): Transparent unit economics, free Developer tier, and Pro/Team plans.
- [MCP Setup](https://actionbox.cloud/mcp): Connect Claude Code, Codex, Cursor, and AI agents through Model Context Protocol.
- [System Status](https://actionbox.cloud/status): Live health metrics for Actionbox Cloud API, push gateways, and webhook queues.
- [Terms of Service](https://actionbox.cloud/terms): Commercial terms, multi-tenant isolation, and service agreements.
- [Privacy Policy](https://actionbox.cloud/privacy): Data protection, tenant privacy, and zero training on private customer payloads.

## Core API Primitives
- [Action Lifecycle](https://actionbox.cloud/docs/actions): Server-owned state machine (`open`, `resolved`, `cancelled`, `expired`).
- [Interaction Types](https://actionbox.cloud/docs/interaction-types): Typed choices, text, numbers, ratings, and forms.
- [Signed Webhook Callbacks](https://actionbox.cloud/docs/webhooks): HMAC-SHA256 authenticated callback delivery.
- [Heartbeat Watches](https://actionbox.cloud/docs/watches): Missed schedules and runtime overruns become ordinary durable Actions.
- [Execution Outcomes](https://actionbox.cloud/docs/decisions-outcomes): Bind one immutable execution result to the exact approved Action.
- [API Reference](https://actionbox.cloud/docs/api-reference): Source-authenticated REST endpoints and typed payloads.
- [MCP Human Input](https://actionbox.cloud/docs/mcp-gate): Connect an MCP-capable agent to durable ask_human and get_action tools through the public CLI bridge.
- [iPhone App Guide](https://actionbox.cloud/docs/mobile-app): Install the app, sign in, configure push preferences, and review Actions safely.

## Blog: Workflows with Human Sign-Off
Copy-paste approval workflows for deployments, data pipelines, cron jobs, incident response, ML model promotion, and secret rotation. Each post includes complete, runnable code.
- [MCP human-in-the-loop: ask a reviewer before an agent acts](https://actionbox.cloud/blog/mcp-human-in-the-loop/): A tested MCP human-in-the-loop example using ask_human and get_action. See how to route a decision to a reviewer and enforce it before another tool runs.
- [TypeSafe Jev Human Review: A Simple ActionBox Example](https://actionbox.cloud/blog/typesafe-jev-human-review/): See a simple TypeSafe Jev human review example: Jev checks a refund, and ActionBox asks a person when the decision is unclear or high-value.
- [TypeSafe AI Jev Review: Tested vs LLMs, API Examples, and Benchmarks](https://actionbox.cloud/blog/typesafe-ai-jev-review/): We tested TypeSafe AI Jev in early access. See Jev vs LLMs, pricing, benchmarks, API limits, and working Python and TypeScript examples.
- [OpenAI Agents API Human Approval with ActionBox](https://actionbox.cloud/blog/openai-agents-api-human-approval/): Route OpenAI Agents API required function calls to ActionBox, collect a typed human decision, and continue the same managed session safely.
- [RPA Exception Handling: Route Failed Bot Runs to Human Review](https://actionbox.cloud/blog/rpa-exception-handling-human-review/): Design RPA exception handling that retries safe failures, pauses at a checkpoint, routes judgment to a human, and resumes without duplicate work.
- [Price Override Approval Workflow for Custom Commerce Systems](https://actionbox.cloud/blog/price-override-approval-workflow/): Build a price override approval workflow with clear discount thresholds, margin context, version checks, expiry, and a safe release path.
- [n8n Workflow Templates: Add Human Approval Before a Risky Step](https://actionbox.cloud/blog/n8n-workflow-templates-human-approval/): Import a tested n8n workflow template that pauses for human approval, branches safely, and reports the result after the approved operation runs.
- [Material Review Board Process: From NCR to Disposition](https://actionbox.cloud/blog/material-review-board-process/): Build a material review board process that moves a nonconformance report from quarantine to an authorized, traceable disposition.
- [Inventory Adjustment Approval Workflow for Cycle Count Variances](https://actionbox.cloud/blog/inventory-adjustment-approval-workflow/): Design an inventory adjustment approval workflow that reviews cycle count evidence, prevents stale writes, and preserves the final outcome.
- [EDI Error Handling: Retry Failures and Route Business Exceptions](https://actionbox.cloud/blog/edi-error-handling-human-review/): Build EDI error handling that retries transport failures, quarantines invalid messages, and routes price or quantity exceptions to a human.
- [Slack Approval Workflow for Developer Operations](https://actionbox.cloud/blog/slack-approval-workflow/): Build a Slack approval workflow for deployments and recovery requests, with clear reviewer access, deadlines, decision records, and execution outcomes.
- [GitHub Actions Slack Notifications: Failures and Approvals](https://actionbox.cloud/blog/github-actions-slack-notification/): Send a Slack notification when a GitHub Actions job fails, link to the run, and separate failure reporting from human approval and recovery.
- [GitHub Actions Schedule Not Running? Diagnose and Monitor It](https://actionbox.cloud/blog/github-actions-schedule-not-running/): Check disabled workflows, the default branch, and scheduling delays. Add independent monitoring for a GitHub Actions workflow that never starts.
- [Cron Job Not Running? Debug It and Detect the Next Missed Run](https://actionbox.cloud/blog/cron-job-not-running/): Diagnose a cron job that works manually but fails on schedule. Check execution, environment, output, and missing-run monitoring.
- [Backup Monitoring: Detect Missing Runs and Validate the Result](https://actionbox.cloud/blog/backup-monitoring/): Monitor scheduled backups without confusing a successful process with a recoverable backup. Separate completion checks from restore tests.
- [Actionable Alerts: From Detection to Verified Recovery](https://actionbox.cloud/blog/actionable-notifications/): Learn what makes actionable alerts useful: a six-field checklist, a failed-import example, and a copyable ActionBox request from detection to recovery.
- [Human-in-the-loop automation: build a human approval workflow](https://actionbox.cloud/blog/human-in-the-loop-automation/): Build a human approval workflow for automation. See a real workflow screenshot and learn how to handle approval, rejection, expiry, retries, and execution.
- [dbt source freshness: configure checks and handle stale data](https://actionbox.cloud/blog/dbt-source-freshness/): Configure dbt source freshness, choose warning and error thresholds, inspect results, and decide what downstream jobs should do when data is late.
- [Data reconciliation: a SQL example and an exception workflow](https://actionbox.cloud/blog/data-reconciliation/): Compare source and target data with a runnable SQL example. Find missing rows and mismatched values, investigate differences, and review exceptions.
- [Data contracts: define expectations and review breaking changes](https://actionbox.cloud/blog/data-contracts/): Define a data contract with schema, meaning, quality rules, and ownership. Separate automated validation from review of breaking changes.
- [Airflow backfill: rerun historical data safely](https://actionbox.cloud/blog/airflow-backfill/): Plan an Airflow backfill, choose reprocessing behavior, limit concurrency, and review historical reruns before they change downstream data.
- [Human-in-the-Loop vs Human-on-the-Loop vs Human-in-Command](https://actionbox.cloud/blog/human-in-the-loop-vs-human-on-the-loop/): Human-in-the-loop vs human-on-the-loop: see the difference, practical workflow examples, and when an AI system needs approval, monitoring, or human-in-command policy.
- [EU AI Act Article 14: Human Oversight Engineering Guide](https://actionbox.cloud/blog/eu-ai-act-human-oversight/): Understand EU AI Act Article 14 human oversight, current deadlines, engineering controls, audit evidence, and ActionBox's implementation boundary.
- [AI Agent Orchestration: A Production Control Loop for Tools, Approvals, and Recovery](https://actionbox.cloud/blog/ai-agent-orchestration/): A reference architecture for AI agent orchestration with durable state, tool policy, human approval, retries, and execution evidence.
- [AI Agent Identity and Access Control: Who Acts, Who Approves, and What Executes](https://actionbox.cloud/blog/ai-agent-identity-access-control/): A practical model for AI agent identity, authentication, authorization, human approval, and least-privilege tool execution.
- [n8n Human in the Loop: Native Review or ActionBox?](https://actionbox.cloud/blog/n8n-human-in-the-loop/): Compare n8n's built-in human review with ActionBox, then build an approval workflow using the community node or the API. Includes a safe first test.
- [CrewAI Human in the Loop Before a Tool Runs](https://actionbox.cloud/blog/crewai-human-in-the-loop/): Add a CrewAI human-in-the-loop approval between an agent proposal and a real side effect, with timeout handling and outcome reporting.
- [MCP Security and Authorization: OAuth, Permissions, and Approval](https://actionbox.cloud/blog/mcp-security-authorization/): Secure an MCP integration by separating OAuth authorization, tool permissions, human approval, secret handling, and auditable execution outcomes.
- [AI Agent Observability: Trace Tool Calls, Decisions, and Outcomes](https://actionbox.cloud/blog/ai-agent-observability/): Design AI agent observability that connects model and tool traces with human decisions, execution outcomes, failures, and rollback evidence.
- [AI Agent Governance: Approval Policies, Audit Trails, and Outcomes](https://actionbox.cloud/blog/ai-agent-governance/): Build practical AI agent governance with risk policies, human approval, version-bound decisions, execution outcomes, and reviewable evidence.
- [Webhook Human Approval Integration: Resume Work Safely](https://actionbox.cloud/blog/webhook-human-approval-integration/): Build and verify a signed Actionbox approval webhook that resumes a deployment exactly once, with tested Python code and a real end-to-end walkthrough.
- [Production AI Agent Deployment: Human Approval Checklist](https://actionbox.cloud/blog/production-ai-agent-deployment-guide/): A production AI agent deployment checklist covering identity, tool permissions, approvals, durable state, observability, and rollback.
- [OpenAI Agents SDK Human-in-the-Loop Approval with Actionbox](https://actionbox.cloud/blog/openai-agents-sdk-human-in-the-loop/): Connect OpenAI Agents SDK human-in-the-loop tool approvals to Actionbox with needs_approval, typed decisions, and fail-closed resume.
- [Adding Human Oversight to LLM Pipelines: A Practical Approval Step](https://actionbox.cloud/blog/llm-human-approval-step/): Add a human approval step to an LLM pipeline before tool calls, customer-facing output, or irreversible automation continues.
- [LangGraph Human-in-the-Loop Approval with Actionbox](https://actionbox.cloud/blog/langgraph-human-in-the-loop/): Connect a LangGraph interrupt to a human reviewer with a stable thread_id, Actionbox approval, and fail-closed Command resume.
- [Human-in-the-Loop API vs Building Your Own Approval System](https://actionbox.cloud/blog/human-in-the-loop-api-vs-build-your-own/): Compare a human-in-the-loop API with an in-house approval system across state, notifications, security, auditability, and long-term maintenance.
- [Durable Execution Explained: Pause and Resume Stateful AI Agents](https://actionbox.cloud/blog/durable-execution-explained/): What durable execution means for stateful AI agents, how checkpoints survive restarts, and where human approval belongs before side effects.
- [Human Approval Tools for AI Agents: 2026 Evaluation Guide](https://actionbox.cloud/blog/best-human-approval-tools-for-ai-agents/): A practical framework for evaluating human approval tools for AI agents: state, safety, typed decisions, audit trails, and developer fit.
- [AI Agent Audit Trails: Approval and Execution Best Practices](https://actionbox.cloud/blog/ai-agent-audit-trail-best-practices/): Build an AI agent audit log that records the proposed tool call, human decision, exact payload, and real execution outcome.
- [Actionbox Review: Human Approval API Alternatives for AI Agents](https://actionbox.cloud/blog/actionbox-vs-alternatives/): An honest Actionbox review and alternatives guide for teams choosing between framework interrupts, workflow engines, alerts, and a human approval API.
- [Actionbox 0.1.4: Watches, Agent Runs, and Outcomes](https://actionbox.cloud/blog/actionbox-0-1-4-autonomy-foundations/): Actionbox 0.1.4 adds heartbeat and progress Watches, immutable execution outcomes, and bounded Agent Run context.
- [Cron Monitoring: Detect Missed Jobs, Failures, and Timeouts](https://actionbox.cloud/blog/heartbeat-watches-that-turn-silence-into-an-action/): Monitor backups, cron jobs, CI runners, and agent tasks with heartbeat Watches that turn missed check-ins into durable Actions.
- [Terraform plan approval: protect destroy and replace operations](https://actionbox.cloud/blog/terraform-destroy-gate/): A copy-paste Terraform plan approval gate that detects destroy and replace operations in CI, blocks apply, and records the human decision.
- [Stripe refund approval workflow: route high-value refunds](https://actionbox.cloud/blog/refund-approval-workflow/): Build a Stripe refund approval workflow that shows a reviewer the amount and reason, checks the human decision, and avoids duplicate refunds on retries.
- [JIT privileged access approval workflow: grant, expire, revoke](https://actionbox.cloud/blog/privileged-access-approval/): A copy-paste JIT privileged access approval workflow with bounded grants, automatic revocation, expiry, and an audit trail without standing access.
- [Policy Exceptions Workflow: Approve, Expire, and Audit](https://actionbox.cloud/blog/policy-exception-approval/): A practical policy exceptions workflow that records approvals, expiry, renewal, owners, and compliance evidence for deviations.
- [GitHub Actions Manual Approval and Deployment Gate](https://actionbox.cloud/blog/github-actions-deploy-gate/): Add a GitHub Actions manual approval step before deployment, compare native environment reviewers with ActionBox, and handle rejection or timeout safely.
- [Flaky test retry approval in CI: quarantine failures safely](https://actionbox.cloud/blog/flaky-test-retry-approval/): A copy-paste flaky test retry approval workflow for CI that limits retries, records the decision, and fails closed when quarantine is rejected.
- [Feature flag rollout approval: gate 5%, 25%, and 100% exposure](https://actionbox.cloud/blog/feature-flag-rollout-approval/): A copy-paste feature flag rollout approval workflow that gates each exposure stage with evidence, guardrails, sequential decisions, and rollback.
- [Dependabot and Renovate approval: gate major dependency updates](https://actionbox.cloud/blog/dependency-pr-approval/): A copy-paste dependency update approval workflow for Dependabot and Renovate that auto-merges safe updates and routes major changes to a human.
- [Database migration approval workflow: block destructive PostgreSQL DDL](https://actionbox.cloud/blog/database-migration-gate/): A copy-paste database migration approval workflow that blocks destructive PostgreSQL DDL in CI with risk detection, lock budgets, and audit history.
- [Argo CD production sync approval: a Kubernetes GitOps gate](https://actionbox.cloud/blog/argocd-approval-gate/): Add an Argo CD approval gate before production sync. Review a specific Git revision, handle rejection and timeouts, and verify the deployment.
- [Human-in-the-Loop AI Agents: Approval Workflow Guide](https://actionbox.cloud/blog/ai-agent-approval-gate/): Build a human-in-the-loop AI agent approval gate: choose which tool calls need review, show the exact proposed action, and resume safely after a decision.
- [dbt full refresh: rebuild models with a reviewed scope](https://actionbox.cloud/blog/dbt-full-refresh-gate/): Use dbt full refresh for incremental models, check configuration overrides, preview model selection, and add approval before a production rebuild.
- [Cron job approval workflow: gate scheduled automation before it acts](https://actionbox.cloud/blog/cron-jobs-that-ask/): A copy-paste cron job approval workflow that gates cleanup, backup, and rotation tasks before they run, with explicit timeout and audit behavior.
- [Incident response approval workflow with an SLA countdown](https://actionbox.cloud/blog/incident-runbook-sla/): A copy-paste incident response approval workflow for rollbacks, scaling, and kill switches with a visible SLA countdown and fail-closed timeout.
- [ML model promotion approval workflow: staging to production safely](https://actionbox.cloud/blog/ml-model-promotion-gate/): A copy-paste ML model promotion approval workflow that gates staging-to-production releases with evaluation metrics and safe rollback behavior.
- [Secret rotation approval workflow: stage, approve, and swap safely](https://actionbox.cloud/blog/secret-rotation-sign-off/): A copy-paste secret rotation approval workflow that stages a new credential, requires approval before production swap, and records audit evidence.