Receipts
Add two headers to a request and its cost is recorded against a workflow and a job. The console and your statements then show cost per finished job, not just tokens per day.
| Header | What it means | Example |
|---|---|---|
X-Continuum-Workflow | The kind of work. Keep it short and stable. | invoice-check |
X-Continuum-Outcome | The one job this request served. Several requests can share it. | INV-2291 |
X-Continuum-Tag-<name> | Anything else you want to group by. Up to 10 per request. | X-Continuum-Tag-Env: production |
Values are trimmed and kept to 100 characters. Tag names may use letters, numbers, - and _, up to 40 characters.
from openai import OpenAI
client = OpenAI(base_url="https://api.continuumlabs.ai/v1", api_key="cl-your-key")
reply = client.chat.completions.create(
model="glm-5.2",
messages=[{"role": "user", "content": invoice_text}],
extra_headers={
"X-Continuum-Workflow": "invoice-check",
"X-Continuum-Outcome": "INV-2291",
},
)curl https://api.continuumlabs.ai/v1/chat/completions \
-H "Authorization: Bearer cl-your-key" \
-H "X-Continuum-Workflow: invoice-check" \
-H "X-Continuum-Outcome: INV-2291" \
-H "Content-Type: application/json" \
-d '{"model": "glm-5.2", "messages": [{"role": "user", "content": "..."}]}'The headers work on chat completions and embeddings, streamed or not.
Agents
An agent may make dozens of calls for one job. Give every call the same X-Continuum-Outcome. The statement counts it as one job, and its cost is the sum of its calls.
Declaring the value first
Before a workflow goes live, declare three things in the console under Cost control:
- What one job is, in a sentence.
- How many minutes it takes a person by hand.
- What would count as failure.
Set the cost of an hour of staff time once for the account. Statements then value each workflow's jobs against what you declared. A return figure set before spending is one a CFO can trust.
What is stored
The header values and token counts are stored. Prompts and answers are not. Don't put personal information in a header value.
