Connect a hosted model.
Add a provider credential for your organization. Requests go to the provider you select.
03 / In production
Check live answers against your task’s rules.
Each request uses a live release and returns a validation result. Review the answer alongside its model, task version, and any sources used.
A standard that stays on the request path.
One request, from answer to review
See how Orlo flags a missing field.
For this support task, every answer needs a category, priority, and escalation decision. The task is set to return a warning when a field is missing.
“Our API has returned errors for twenty minutes. Every customer request is failing.”
SUPPORT-042Example data
Returned with a warningRelease v04
Priority is missing.
- Category
- API outage
- Priority
- Not returned
- Escalate
- Yes
The required-fields rule flagged this answer. With warning behavior, the application receives the answer and the warning.
What stays in the record
The input, returned answer, model, task version, release, and validation result stay together. When the task uses documents, you can also inspect the retrieved passages.
Explore the API
The Orlo workspace
Connect your application to a live release.
Use the OpenAI-compatible endpoint to call the task. Each answer includes a validation status your application can act on.
Connect something new
Connect your application to the task’s live release.
OpenAI-compatible endpoint
TypeScript
Python and LangGraph
For task
Support ticket routing ⌄
Keep your OpenAI client. Point it at Orlo and add the task header. Every answer comes back with an X-Orlo-Validation header.
from openai import OpenAI
client = OpenAI(
base_url=ORLO_GATEWAY_URL,
api_key=ORLO_API_KEY,
)
client.chat.completions.create(
model="claude-sonnet-4-20250514",
messages=messages,
extra_headers={"X-Orlo-Task-Id": ORLO_TASK_ID},
)Choose where your models and Orlo run.
Your model endpoint and your Orlo installation are separate choices. Decide where each belongs.
Use your own model endpoint.
Connect an endpoint you operate, including vLLM or llama.cpp. Compare it using the same task examples.
Host Orlo in your environment.
Run Orlo on infrastructure your team manages. Your application still executes the work and enforces agent permission decisions.