Meltbox Guide Docs Start free →

Landscape · verified July 2026

HumanLayer alternatives (2026)

If you came looking for HumanLayer's human-in-the-loop approval SDK, here's the honest situation: it's deprecated, and HumanLayer itself has been rebuilt as an AI coding IDE. So the real question is what to use instead. This is the sourced landscape — including where Meltbox fits and where it doesn't.

Create a free workspace Meltbox vs. HumanLayer

Why you're here

What happened to HumanLayer

HumanLayer's original open-source SDK let you gate an agent's tool calls behind a human approval over Slack or email. That code is now marked "pretty much all deprecated" in its own repository, which points to a rebuilt commercial product.¹ The rebuilt product — still branded HumanLayer — is an AI coding IDE, not a drop-in approval SDK.² So if you were using the SDK to keep a human in the loop, you're effectively shopping for a replacement. Here are the real options.

The options

Five honest replacements

1 · HumanLayer itself — now an AI coding IDE

If you liked HumanLayer and want the team's current direction, their rebuilt product is an IDE for orchestrating coding agents: multi-repo workspaces, a plan-then-implement workflow, and human review of plans before implementation. Free tier (3 seats, 200 sessions/month), Pro $100/user/month.²

Best for: teams who want an opinionated editor. Not: a lightweight approval gate you drop into an existing agent.

2 · gotoHuman — an approval inbox for agent workflows

gotoHuman ("AI Agents unlocked with human supervision") is the closest surviving "review-and-approve" product: build a review template, send a request through an integration, review items in an Agent Inbox or via email/Slack notifications, and a webhook returns the decision to continue the workflow. Integrations include n8n, Make.com, a TypeScript and Python SDK, LangGraph, an MCP server, and an HTTP API.³

Best for: automation/workflow builders (n8n/Make/LangGraph) who want structured review forms. Note: oriented at workflow platforms rather than coding-agent operators specifically; no secret proxy.

3 · LangGraph Agent Inbox — if you build in LangGraph

LangGraph has human-in-the-loop as a first-class primitive: interrupt() pauses a graph and saves state via a checkpointer; you resume with Command(resume=...). The open-source Agent Inbox is a UI for reviewing those interrupts. It's the native, honest choice — if your agent is a LangGraph graph you're willing to build and host.⁴

Best for: teams already on LangGraph. Not: for people running Claude Code / Codex who don't want to rebuild their agent as a graph.

4 · Roll your own — a Slack webhook + a gate

The DIY route: an approval layer that pauses the agent and posts a Slack (or ntfy) message, resuming on your reply. It's real and free, and fine for one agent. The costs are the usual ones — you maintain the glue, a ping isn't a decision surface, and there's no secrets story. We cost this out in Meltbox vs. building it yourself.

Best for: a single agent, one machine. Not: a fleet, or anything touching shared secrets.

5 · Meltbox — a control plane for coding-agent operators

Meltbox comes at it from the operator's side rather than the framework's. Any agent that can run a shell or speak MCP connects; it pushes briefs — clickable decision pages — to one keyboard-fast inbox, and you answer from your desk or your phone (installed PWA push). It also brings a Secret Proxy so agents use your keys without seeing them, which none of the pure approval tools do. It is agent-agnostic and not an IDE.⁵

Best for: people running Claude Code / Codex / Cursor who want decisions, secrets, and a phone push in one place, without switching editors or rebuilding their agent. Not: a framework primitive you embed in graph code — it's a product you connect an agent to.

At a glance

How they line up

What it isBest forSecrets for agents
HumanLayer (rebuilt IDE)AI coding IDE²Teams wanting an editorNot stated
gotoHumanApproval inbox + SDKs³n8n / Make / LangGraph flowsNo
LangGraph Agent InboxFramework primitive + UI⁴LangGraph buildersNo
Roll your ownSlack/ntfy + a gateOne agent, one machineNo
MeltboxControl plane (briefs + proxy)⁵Claude Code / Codex operatorsYes — Secret Proxy

FAQ

Common questions

Is the HumanLayer SDK gone?

The original open-source approval SDK is marked deprecated in its own repository, which points to a rebuilt commercial product. That rebuilt product — still branded HumanLayer — is an AI coding IDE rather than a drop-in approval SDK, so existing SDK users are effectively looking for a replacement.

What's the closest replacement for gating agent actions behind human approval?

It depends on your stack. gotoHuman is an approval-inbox product with SDKs, an MCP server, and n8n/Make/LangGraph integrations. If you build in LangGraph, its native interrupt() plus the Agent Inbox UI covers it. For coding-agent operators who want decisions, secrets, and a phone push in one place, Meltbox is the control-plane option.

Which is best for someone running Claude Code specifically?

HumanLayer's own rebuilt product is an IDE built around orchestrating Claude Code. Meltbox is agent-agnostic: any agent that can run a shell or speak MCP connects, pushes briefs you triage, and uses your secrets through the proxy — without you switching editors.

The control-plane option, free to try

Create a free workspace Watch an agent say hello

Or hand this to your own agent right now:

Read https://docs.meltbox.ai/quickstart and https://docs.meltbox.ai/invite-an-agent, then walk me through: creating a free Meltbox workspace at meltbox.ai, minting an invite link, redeeming it, and pushing me a first brief with a real decision about our project — then watch it for my reply.

More guides

Meltbox vs. HumanLayer →The head-to-head, after the pivot. Approve agent actions from your phone →Human-in-the-loop, on mobile. Meltbox vs. building it yourself →The roll-your-own route, costed. Zero-inbox for agent output →Triage many agents like email.

Sources