TechnoMeta
Talk to us
For operations and finance heads File-based and legacy systems

The manual work between files and systems, done for you.

Every day someone opens a scanned invoice, checks it against the purchase order and types it into the ERP. Workflow Agents watch the folders, mailboxes and exports where that work starts, carry it through the interfaces the system already has, and stop for a person whenever a step would change money or records. That person sees what the software found and approves, corrects or rejects it with one click.

A Workflow Agents run paused at the approval gate. Sample data.

What it does

Starts where work starts

Watched folders, mailboxes, scheduled exports or database events. Nothing changes about how files arrive.

Reads and classifies files

Each invoice, challan or form is recognised and its fields extracted, with a confidence score on each.

Matches your records

Purchase orders, deliveries, account codes: the agent checks the file against what the system already holds.

Uses existing interfaces

SOAP and REST services, stored procedures, file imports. Your legacy system does not need to change.

A person approves writes

Posting to the ERP waits for a named approver, who can approve, correct or reject with one click.

Every run can be replayed

Each step is logged with its time and result, so a run can be reviewed afterwards or replayed after a fix.

How a Workflow Agents pilot works

A fixed-price Discovery first, then a fixed-price pilot on one workflow. The three stages are described on the homepage; this is what they mean for Workflow Agents.

  • You give

    One workflow, a month of its files, access to the system's own interfaces, and the person who does the work today.

  • We measure

    Fields extracted and matches made correctly, minutes per document before and after, and a logged approval on every write.

  • You get

    The workflow running on your real files, with the approval screen, the run log and a report against those numbers.

Pilot price fixed in writing at the end of Discovery.

Google Cloud services it uses

ServiceUsed forWhy
Cloud Run jobsOne job per runScales with the queue and to zero when it is empty.
Pub/Sub and Cloud TasksQueues and retriesEach run is a queued message with retries and a durable record.
Gemini on Vertex AIReads, classifies and calls toolsFunction calling keeps each action explicit and logged.
Secret ManagerERP and file-share credentialsCredentials stay in your project; the model is given none.
Cloud SQLRun state and approvalsWho approved what, and when, as a transactional record.

Most of the load is Gemini function calling on Vertex AI for each run and Cloud Run jobs that scale with the queue. The deployment is its own Google Cloud project, in the region you choose.

On your side: a connector in a container or VM that holds the ERP and file-share credentials, makes outbound TLS connections only, and can be switched off by your IT team at any moment.

Security and data notes

A person approves writes

Agents read, match and propose. Posting to the ERP, changing a record or sending a document waits for a named approver, and the approval is part of the audit trail.

Credentials stay with you

ERP and file-share passwords live on the connector or in Secret Manager in your project. The model sees only the fields a task needs.

Every run logged

Each step, its time and its result are logged in your project and can be exported to your SIEM. Runs can be replayed after a fix.

Questions operations teams ask

Can the agents write to our ERP?

Yes, through the interfaces the ERP already has: SOAP or REST services, stored procedures, file imports. Every write waits for a named approver, is logged, and is limited to the workflow in the pilot plan. Read-only is the default until you grant more.

What happens when the agent is unsure?

It stops and asks. A low-confidence field, a match that does not line up, or an amount outside the agreed range all pause the run for a person, who sees what was found and decides. Nothing is posted while a run is paused.

Do we have to change our legacy system?

No. The agent works through what the system already exposes: SOAP or REST services, stored procedures, scheduled file imports. Where the only way in is a file import, the agent writes the file and a person approves it before it is picked up.

How is a run checked afterwards?

Every step is logged with its time, its input and its result, in your project. A supervisor can open any run, see what the agent read and did, and replay it after a correction.

Talk to us about Workflow Agents

Tell us which workflow, which system it ends in, and how many documents a month. The form is on the homepage, or write to contact@technometa.com.

Go to the contact form