Introducing Relayna: Human-in-the-Loop for AI Workflows

AI agents are getting remarkably capable. They can write code, draft documents, browse the web, and orchestrate complex multi-step workflows — often without any human guidance at all.

But there’s a category of tasks where autonomy isn’t enough. When an agent is about to send an email to a client, transfer funds, publish content, or hand off a sensitive file, you need a human to review and approve it first. That’s the gap Relayna fills.

What Relayna Does

Relayna is a secure file relay and human-approval layer for AI workflows. At its core, it gives your agents a simple REST API to:

1. Upload files — anything from PDFs and images to CSVs and JSON
2. Create a review checkpoint — a structured request for human approval
3. Generate a magic review link — a token-based URL you can send to any reviewer, no account required
4. Wait for a decision — via webhook callbacks or lightweight status polling

The reviewer sees a polished UI showing the files and any context your agent provided. They can approve, reject, or request changes. Once they decide, Relayna fires a webhook back to your system so the agent can continue.

Why We Built It

We kept running into the same pain point while building AI-assisted workflows. The agent part was easy — the hard part was the handoff. You’d end up with a mess of ad-hoc email threads, shared drives, and manual status checks.

We wanted something that felt like a first-class primitive. Something an agent could call as naturally as any other API endpoint, that would handle the entire review lifecycle automatically.

How It Works

Here’s a minimal example using our Python client:

import relayna

client = relayna.Client(api_key="your-api-key")

# Upload the file the agent generated
asset = client.assets.upload("report.pdf")

# Create a review checkpoint
checkpoint = client.checkpoints.create(
    title="Q1 Financial Report — Please Review",
    asset_ids=[asset.id],
    webhook_url="https://your-app.com/webhooks/relayna",
    ttl_seconds=86400  # 24-hour expiry
)

# Send the review link to your team
print(checkpoint.review_url)

Your webhook receives a POST when the reviewer decides, with the full decision payload including any notes they left.

What’s Next

We’re just getting started. In the coming months we’re shipping:

• Structured decision fields — let reviewers fill in forms, not just approve/reject
• Multi-reviewer workflows — require consensus from multiple people before proceeding
• Audit logs — a full trail of every review decision for compliance
• SDK libraries — official clients for Python, TypeScript, and Go

We’d love to hear how you’re building with AI agents and where human review is still a bottleneck for you. Reach out at [email protected].