AskHuman by thrain.ai

The control layer that makes AI agents deployable.

Your agents could reorder stock, issue refunds, and reprice listings today. What's stopping you isn't capability — it's accountability. AskHuman gates every consequential agent action behind real human sign-off: delivered to your team's phones, answered in one tap, logged forever.

1 · The agent proposes
When an agent hits an action that moves money, inventory, or customers, it calls ask_human instead of acting alone. One HTTP call, or the MCP tool.

2 · Your people decide
The decision lands in Slack, Discord, or any webhook as a one-tap answer link that works on any phone. No new app, no login, no training.

3 · The action proceeds — signed
The agent gets the decision and acts on it. Every agent-initiated action now carries a human signature, a rationale, and a timestamp.

What it looks like in your ops

POST /v1/asks
{
  "question": "Branch 12 stocks out of SKU-4471 in ~6 days.
               Order 400 units from Meridian Supply ($8,460)?",
  "type": "choose",
  "options": ["Order 400", "Order 200 (conservative)", "Skip — seasonal dip"],
  "sla_seconds": 14400
}

→ (your ops lead taps "Order 400" from their phone)
→ { "status": "answered",
    "answer": { "choice": "Order 400", "rationale": "velocity is real, not promo-driven" },
    "answered_by": "D. Alvarez" }

Purchase orders. Refunds and credits. Price changes. Customer-facing messages. Vendor actions. Production config. If an agent shouldn't do it unsupervised, it goes through AskHuman.

The deployability gap

Every company has agent pilots. Almost none have agents in production touching money — because the blocker was never the model, it's that nobody will sign off on unsupervised spend. AskHuman is the missing sign-off. It turns "too risky to deploy" into "approved, audited, shipped" — and it means the day an auditor asks "which of these actions did a human approve?", the answer is a query, not an investigation.

Who it's for

Ops, finance, and platform teams putting agents to work on real business processes — and the leaders who have to answer for what those agents do. Human oversight of automated decisions is becoming a requirement, not a preference. Be able to prove yours.

Early access — free

AskHuman is free: MIT-licensed (source on GitHub) with a hosted API and MCP server. We onboard teams by hand while in early access.