Skip to main content

FinOps

Put a clear owner on every governed call. Trinitite records the model cost, then groups it by scope, team, project, or cost center.

Status: Beta Companion docs: Models and routing · Scopes and assets · SDK observability Source of truth: Analytics API · Proxy API.

Label the work once​

Put the cost label on a scope. Every governed call from a bound asset carries it:

refunds = tr.scope(
    "refund-flow",
    goal="Resolve valid refunds without exposing customer data.",
    cost_center="cs-dept-829",
    tags={
        "team": "support",
        "project": "q3-refunds",
        "env": "prod",
    },
)

Read cost on the result​

response = refunds.client("openai").chat.completions.create(
    model="gpt-4o",
    messages=messages,
)
result = tr.result(response)

print(result.cost_center)
print(result.cost_usd)
print(result.routed_model)

cost_usd can be absent when the provider does not return enough usage data. Check it before doing per-call calculations.

See spend for one scope​

summary = tr.finops.spend(
    scope="refund-flow",
    since="30d",
)

print(summary.total_spend_usd)
print(summary.call_count)
print(summary.estimated)

The summary separates input, output, and reasoning cost when those fields are available.

Build a chargeback report​

report = tr.finops.chargeback(
    group_by="cost_center",
    from_="2026-08-01",
    to_="2026-08-31",
)

for row in report.rows:
    print(row.key, row.total_spend_usd, row.call_count)

Group by cost_center, scope, tag, nhi, model, or provider.

Drill into one cost center by model:

breakdown = tr.finops.breakdown(
    by="model",
    cost_center="cs-dept-829",
)

Keep labels and limits separate​

A cost center answers who owns the spend. A spend limit stops an agent identity from running past a budget.

support_agent = tr.agent(
    "support-bot",
    goal="Resolve support requests.",
    spend_limit_usd=500,
)

status = tr.finops.breaker_status(nhi=support_agent.identity.id)
print(status.current_spend_usd)
print(status.spend_limit_usd)
print(status.breaker_tripped)

Use both when the team needs chargeback and a hard stop.

Lower cost with an explicit route policy​

refunds = tr.scope(
    "refund-flow",
    goal="Resolve valid refunds.",
    cost_center="cs-dept-829",
    route="auto",
    cost_ceiling_usd_per_call=0.25,
)

The router records the selected model and its reason on the result, so savings stay explainable.

Where to go next​

Last updated on