The best HumanLayer alternative for human-in-the-loop AI agents

HumanLayer was one of the first products built for human-in-the-loop AI agents. Its SDK let a function be wrapped with an approval decorator so that, when an agent called it, a person was asked in Slack or by email before it ran. Plenty of teams built on that idea.

In 2026 the company changed direction. Its website now sells a multiplayer coding agent IDE and cloud, and the README of its original GitHub repository says the code there is “pretty much all deprecated”. If you came looking for agent approvals, you need another option.

What HumanLayer did

• Wrapped functions so they required approval before an agent could run them.
• Let an agent ask a person a free-form question (“human as a tool”).
• Delivered requests over Slack, email and other chat channels, and returned the answer to the agent.

Those are still exactly the right primitives. The question is who provides them now.

The closest replacement: redphonr

redphonr covers the same ground and adds what HumanLayer users often asked for.

	HumanLayer (original SDK)	redphonr
Current focus	Coding agent IDE	Human approvals for agents
Integration	Python and TypeScript SDK decorators	Native MCP server, plus HTTP API
Approve or deny	Yes	Yes (request_approval)
Ask a question	Yes	Yes (ask_human, collect_input)
Review output before sending	Build it	Yes (verify_output)
Channels	Slack, email, chat	Slack, SMS, WhatsApp, Telegram, Discord, email, voice
Escalation	Not in its current docs	Ladders, on-call and backup, weekly rotation
Timeout default	Not in its current docs	Deny by default

The biggest change is the integration model. Instead of decorating functions in one SDK, redphonr is an MCP server, so any MCP-capable agent (Claude, Cursor, the OpenAI Agents SDK, LangChain, CrewAI, Mastra) can call it without a library.

Migrating a require_approval function

Before, with a decorator:

@hl.require_approval()
def send_invoice_reminders(customer_ids: list[str]) -> str:
    ...

After, with an explicit gate:

async def send_invoice_reminders(session, customer_ids: list[str]) -> str:
    result = await session.call_tool("request_approval", {
        "message": f"Send payment reminders to {len(customer_ids)} customers?",
        "timeout_seconds": 3600,
        "default": "deny",
    })
    if result["decision"] != "approved":
        return f"Not sent: {result.get('note') or 'not approved'}"
    ...

Or, if your agent already speaks MCP, add the server and let the model decide when to ask, guided by your system prompt. The 5-minute tutorial shows the prompt.

Other alternatives

• gotoHuman if your approvals are mostly reviewing AI drafts in a web inbox. See redphonr vs gotoHuman.
• Your framework’s own pause (LangGraph interrupt(), OpenAI Agents SDK needs_approval) if the approver is already in your app’s UI.
• The full field is in the 10 best human-in-the-loop tools for AI agents.