Teams with a shared inbox or helpdesk where every message has to be read, sorted and passed on.
What it costs today
Someone triages the queue by hand: reads each message, works out what it's about, copies the order number into another tool and pings the right person.
What solved looks like
Each message is classified and its details extracted the moment it arrives; billing questions get a drafted reply, everything else lands in the right Slack channel.
Every message labelled and its key fields pulled out, consistently.
The right people notified with the details already filled in.
A run log for every message: each step's output, and which step failed and why.
How it runs
Inbox automations step by step
Each stage is a real step in SentientOne. Timings are medians from our latest benchmark; your model and network change them.
Inbox automation flow, a published workflow. Your helpdesk posts the message to the workflow's webhook URL with its token. 1: the Start block checks the input fields. 2: Classify sorts the message into billing, bug or other. 3: Extract pulls out the order id, product and severity. 4: a Condition checks whether the label is billing. 5: if so, an Agent block drafts a reply from your knowledge. 6: Send email sends it; other messages go to a Slack message block instead. The run finishes within 280 seconds and at most 25 blocks.
Trigger
Your helpdesk's webhook → the workflow's Incoming webhook URL
Payload
“{"email": "sam@example.com", "message": "I was charged twice for order 10482"}”
Headers
X-Workflow-Token: wft_…
?wait=true to wait up to 30 s
60 calls a minute, 100 KB body
SENTIENTONE
Start
Webhook
The webhook's JSON body becomes the run's input, checked against your fields.
Published snapshot only
Classify
1 credit
Sorts the message into labels you describe; each label is its own exit.
billing · bug · other
Extract
1 credit
Pulls typed fields out of the text for the next steps.
order_id · product · severity
Condition
Free
Billing goes to the agent; everything else goes to Slack.
label equals billing
Agent
1 credit
Your support agent drafts a reply from your billing policies.
Customer Support agent
Send email
Outbound
Sends the draft to the customer (or post it to Slack for review).
≤ 5 recipients · 100 a day
Run log
Succeeded · 4 credits
classify_1 → billing. extract_1 → order_id 10482, product "Annual plan", severity medium. condition_1 → true. agent_1 drafted a refund reply. send_email_1 → sent to sam@example.com.
Agents run on your own model provider key; there's no shared one, so nothing answers until you add it. Create a key with OpenAI, Anthropic, Google Gemini or Groq, then open Configure → LLM keys → Add key. It's encrypted at rest with AES-256-GCM, used only to call your provider, and never shown again in full. If a reply fails with INTERNAL_ERROR straight away, the agent has no key; with LLM_AUTH_ERROR, the provider rejected it.
Agents → New agent → Start from: Customer Support, and upload your billing policy (a 6-page PDF is 30 credits) on the Knowledge tab. The workflow will call it.
Workflows → New workflow. In the Start block define the fields email and message. Add Classify (labels billing, bug, other), Extract (order_id, product, severity), a Condition on the label, an Agent block with your agent, and Send email; connect the false branch to a Slack message block.
In the Start block's Webhook trigger, Create token and copy the Incoming webhook URL. Configure your helpdesk or inbox to POST each new message there.
What your helpdesk sends
curl-X POST "https://app.sentientone.ai/api/workflows/hooks/$WORKFLOW_ID" \-H"Content-Type: application/json" \-H"X-Workflow-Token: $WORKFLOW_TOKEN" \-d'{"email": "sam@example.com", "message": "I was charged twice for order 10482"}'
The designer's runs list keeps the last 50 runs with each step's input, output and timing; failed runs say which step failed and why. Turn on auto top-up under Billing.