Build Agentic Workflows That Verify Every Step

Your AI agent can search, analyze, and generate. But can it prove its answers?

Most agent pipelines chain API calls sequentially: fetch data, summarize, deliver. The output looks polished, but nothing in the chain validates whether the data was accurate, the summary was faithful, or the conclusion holds up against contradicting evidence.

Agentic workflows are different. Each step produces structured output — with confidence scores, evidence chains, and contradiction detection — and the next step only runs if the previous one passes verification.

What Makes a Workflow “Agentic”

Chaining three API calls isn't a workflow. It's a script. An agentic workflow has three properties that scripts don't:

Example: Earnings Watcher Workflow

Here's a real workflow that monitors earnings season. Four steps, each one verified:

If Step 3 finds contradictions — say analyst targets diverge sharply from sentiment signals — the workflow flags it in the output instead of silently passing bad data downstream.

Build It in Python

The VEROQ Python SDK handles pipeline orchestration, verification gates, and delivery. Here's the Earnings Watcher as code:

from veroq import VeroqClient

client = VeroqClient(api_key="your-api-key")

# Define the workflow pipeline
workflow = client.workflows.create(
    name="Earnings Watcher",
    steps=[
        {
            "endpoint": "/events/calendar",
            "params": {"days": 7, "type": "earnings"},
            "output": "upcoming_earnings"
        },
        {
            "endpoint": "/ticker/{{ticker}}/score",
            "for_each": "upcoming_earnings.events",
            "output": "scores"
        },
        {
            "endpoint": "/verify",
            "body": {
                "claims": "{{scores.summary}}",
                "mode": "both"
            },
            "gate": {"min_confidence": 0.7},
            "output": "verified_scores"
        },
        {
            "endpoint": "/generate",
            "body": {
                "prompt": "Pre-earnings briefing for {{verified_scores}}",
                "format": "structured"
            },
            "output": "briefing"
        }
    ],
    delivery={"type": "slack", "channel": "#earnings-watch"},
    schedule="0 8 * * 1-5"  # Weekdays at 8am
)

The gate on Step 3 is the key. If verified confidence drops below 0.7, the workflow halts and notifies you instead of generating a brief from unverified data.

Scheduling and Triggers

Workflows can run on a schedule (cron syntax) or trigger on events:

# Run every weekday at market open
workflow.schedule("0 9 * * 1-5")

# Or trigger on specific events
workflow.on_trigger(
    event="watchlist_match",
    watchlist_id="wl_abc123",
    min_confidence=0.8
)

Scheduled workflows run on VEROQ infrastructure — no cron jobs to maintain, no servers to keep alive. Results are delivered automatically.

Delivery: Results Where You Work

Every workflow needs a destination. VEROQ supports four delivery targets:

Delivery targets are configured once in your dashboard settings and reused across all workflows. Add a Slack workspace, and every workflow can deliver to any channel.

Other Workflow Ideas

Build Your Own

Every workflow is a pipeline of VEROQ endpoints with verification gates between them. You pick the steps, set confidence thresholds, choose delivery targets, and schedule when it runs.

• Workflow Builder — create and manage workflows visually
• API Reference — all available endpoints for pipeline steps
• Python SDK — build workflows in code