Framework-agnostic approval API

The pause button for AI agents.

Your agent hits a high-stakes decision. It pauses. A human decides. It resumes. Three lines of code. Zero infrastructure.

// That's it. That's the integration.
const decision = await holdon.ask({
  action: "Send $4,200 wire to vendor",
  context: "Invoice #8821, verified by AP agent",
  notify: ["telegram", "email"]
});

if (decision.approved) executeWire();
How it works

Four steps. One API call.

1

Agent POSTs an approval request

Full context attached. What action, why, what's at stake. Your agent keeps working on other tasks.

2

Human gets notified

Telegram, Email, Slack, or WhatsApp. Wherever they are. Rich context, not a cryptic alert.

3

One tap to decide

Approve or reject. No login required. No context switching. Decision in seconds.

4

Agent resumes with the answer

Webhook fires. Agent picks up exactly where it left off. Full audit trail logged.

14 days
Average time teams spend building their own approval flow
83%
Of enterprise AI deployments need human-in-the-loop
3 lines
Of code to integrate HoldOn into any agent
Why HoldOn

Every framework has HITL. None of them are portable.

</>

Framework-agnostic

Works with LangChain, CrewAI, AutoGen, OpenAI, Claude, or your custom stack. One API for all of them.

 ⇆

Non-blocking

Agent keeps working while waiting for approval. No frozen processes, no timeouts, no wasted compute.

☷

Full audit trail

Every decision logged. Who approved, when, with what context. EU AI Act compliant from day one.

⚡

Multi-channel

Telegram, Email, Slack, WhatsApp. Meet reviewers where they already are. One-tap decisions.

Stop rebuilding approval flows. Start shipping agents.

HoldOn is the missing primitive between autonomous AI and human judgment. Every agent framework solved reasoning. None of them solved the pause.

Read the Docs