SAYDO: voice-driven enterprise workflow automation, powered by Vocal Bridge

Upload a workflow diagram, confirm the business rules, and run the whole process by speaking, with legacy systems like SAP and AS/400 handled underneath.

The problem: business knows the workflow, IT owns the systems

In most enterprises the people who understand a business process are not the people who can wire it up. The business team designs the Invoice-to-Payment flow, the approval tiers, the variance rules. The IT team owns the connections to the systems where that work actually happens, AS/400, SAP, Salesforce, Excel.

That split is slow and lossy. When business wants a new workflow, they hand it to IT, IT spends time decoding it, and even then the legacy-system knowledge needed to run it cleanly is rare. The result is a backlog where a process everyone agrees on sits waiting for someone who can translate it into the language of the systems underneath.

SAYDO's bet is that the business user should be able to describe the process and then run it, without a round trip through engineering.

What Gaurav built

SAYDO collapses designing a workflow and running it into one conversation. A business user uploads an image of a process diagram and SAYDO reads it, then asks the questions a real implementation needs answered, like approval thresholds, sign-off limits, acceptable variances, and which system to trust for a lookup. The user answers with a word or a click, and SAYDO loads the configured workflow onto a canvas.

The point is what happens next. Once a one-time mapping links technical field names to human-readable business terms, the user runs the whole multi-system process by speaking, naming the workflow and the record while the canvas highlights each step as the agent walks the logic. They never open the underlying systems. They say what they want and watch it move. Voice fits because business teams know the workflow but not the legacy software, and talking lets them run it without IT.

This is an early build. The voice execution in the demo includes some rough edges, and the dictionary mapping is a manual setup step that Gupta calls out himself. What it demonstrates is the shape of the idea, a business user operating a real multi-system process by talking to it.

How SAYDO uses Vocal Bridge

SAYDO is built on Vocal Bridge's Voice for Agent surface. The automation agent is the autonomous part. It parses the uploaded diagram, holds the configured workflow, evaluates each decision node, and reaches into the connected systems. Vocal Bridge layers the live voice channel onto that agent so a person can drive it and hear back from it in real time.

The architectural split is clean. SAYDO owns the workflow engine, the diagram parsing, the dictionary that normalizes system terms to business language, and the connectors to SAP, AS/400, Salesforce, and Excel. Vocal Bridge owns the voice path, the real-time transport, streaming speech-to-text and text-to-speech, barge-in, and the sub-second turnaround that lets the canvas light up while the user is still talking. Gupta did not have to build WebRTC plumbing or a streaming voice loop to make "say execute, watch it run" feel immediate.

from vocalbridge import Agent, tool

agent = Agent(name="saydo-runner")

@tool
def execute_workflow(workflow_id: str, invoice_number: str) -> dict:
    """Run a configured workflow node by node and stream node state to the UI."""
    wf = load_workflow(workflow_id)
    for node in wf.walk(context={"invoice_number": invoice_number}):
        agent.emit("node_state", {"id": node.id, "status": node.status})
    return {"workflow_id": workflow_id, "result": wf.outcome}

@tool
def reject_workflow(workflow_id: str) -> dict:
    """Follow the rejection branch from the active decision node."""
    return load_workflow(workflow_id).take_branch("no")

agent.serve()  # Vocal Bridge handles the live voice channel

Illustrative. The actual SAYDO implementation parses richer diagrams and connects to live enterprise systems.

What's hard about voice here

The agent has to interpret a spoken command against the live state of a running workflow, not in the abstract. "Reject the workflow" only means something next to the decision node the agent is sitting on, so the voice turn has to be tied to the workflow's current state and the canvas kept in step with the speech. Vocal Bridge's low-latency channel is what keeps that visual feedback feeling like a response rather than a delay.

Keeping speech and a running multi-system process in lockstep is the hard part, and the real-time layer underneath it is what Vocal Bridge handles.

"Once the one-time mapping is done, the business person can simply say 'execute' and it will execute." Gaurav Gupta, creator of SAYDO