Control Costs with the Analytics API

The activity API gives you daily visibility into your token usage and DA spend. This recipe shows how to query the API, detect spending spikes, and build automated cost controls.

Endpoints#

Both endpoints are key-authed inference routes (Authorization: Bearer $OPENDUNES_API_KEY):

Endpoint	Description
`GET /v1/activity`	Daily usage rolled up per model — tokens, requests, DA spend
`GET /v1/credits`	Lifetime DA totals — deposited and spent

For request-level logs and CSV/JSON export, use Dashboard → Analytics or the activity export.

Checking current spend#

Shell

# Last 30 days, one row per day and model
curl "https://opendunes.com/v1/activity" \
  -H "Authorization: Bearer $OPENDUNES_API_KEY"

JSON

{
  "data": [
    {
      "date": "2026-06-30",
      "model": "anthropic/claude-opus-4.8",
      "model_permaslug": "anthropic/claude-opus-4.8",
      "endpoint_id": "a223cc6b-19a9-48ac-a297-3210a86073e5",
      "provider_name": "anthropic",
      "prompt_tokens": 482000,
      "completion_tokens": 123000,
      "requests": 1245,
      "usage": 184.23
    }
  ]
}

usage is the DA spent that day on that model. Pass ?date=YYYY-MM-DD to narrow to a single day.

Spend by model#

The rows arrive grouped per day × model, so a per-model breakdown is a small aggregation:

Python

import os, requests
from collections import defaultdict

BASE    = "https://opendunes.com"
HEADERS = {"Authorization": f"Bearer {os.environ['OPENDUNES_API_KEY']}"}

rows = requests.get(f"{BASE}/v1/activity", headers=HEADERS).json()["data"]

by_model = defaultdict(float)
for r in rows:
    by_model[r["model"]] += r["usage"]

for model, da in sorted(by_model.items(), key=lambda x: -x[1]):
    print(f"{model}: {da:.2f} DA")

This breakdown tells you which models are driving most of your spend — often the starting point for optimization.

Building a daily spend alert#

Python

def check_daily_spend(threshold_da: float) -> None:
    from datetime import date

    resp = requests.get(
        f"{BASE}/v1/activity",
        params={"date": date.today().isoformat()},
        headers=HEADERS,
    )
    resp.raise_for_status()

    total = sum(r["usage"] for r in resp.json()["data"])
    if total > threshold_da:
        send_alert(f"Spend spike: {total:.2f} DA so far today")

def send_alert(message: str) -> None:
    # Send to Slack, email, PagerDuty, etc.
    print(f"ALERT: {message}")

# Run this on a cron every hour
check_daily_spend(threshold_da=100.0)

Checking remaining balance#

Shell

curl https://opendunes.com/v1/credits \
  -H "Authorization: Bearer $OPENDUNES_API_KEY"

JSON

{
  "data": {
    "total_credits": 5000,
    "total_usage": 4575
  }
}

The remaining balance is total_credits - total_usage (DA). The X-Balance-Available header on every inference response also returns the current balance in micro-DA — useful for inline checks without a separate API call.

Automated low-balance protection#

Python

def balance_da() -> float:
    r = requests.get(f"{BASE}/v1/credits", headers=HEADERS)
    r.raise_for_status()
    data = r.json()["data"]
    return data["total_credits"] - data["total_usage"]

LOW_BALANCE_THRESHOLD_DA = 50.0

if balance_da() < LOW_BALANCE_THRESHOLD_DA:
    # Pause job queue, alert on-call, redirect to a cheaper model tier
    pause_inference_queue()
    send_alert(f"Balance below {LOW_BALANCE_THRESHOLD_DA} DA — topping up required.")

Cost optimization checklist#

Model downsizing — from the by-model breakdown, identify high-spend models and test whether a cheaper model meets your quality bar on those tasks.
Prompt compression — long system prompts multiply cost across every request. Trim where possible.
max_tokens caps — set a realistic max_tokens per request type; runaway generations can multiply expected cost by 5–10×.
Rate-limit tuning — if a batch job is generating thousands of requests in a short window, adding a delay reduces peak spend without affecting throughput.
Cheap models for triage — sort the catalog by price and use a low-cost model for classification and routing decisions that don't require a premium one.

The activity API gives you daily visibility into your token usage and DA spend. This recipe shows how to query the API, detect spending spikes, and build automated cost controls.

Endpoints#

Both endpoints are key-authed inference routes (Authorization: Bearer $OPENDUNES_API_KEY):

Endpoint	Description
`GET /v1/activity`	Daily usage rolled up per model — tokens, requests, DA spend
`GET /v1/credits`	Lifetime DA totals — deposited and spent

For request-level logs and CSV/JSON export, use Dashboard → Analytics or the activity export.

Checking current spend#

Shell

# Last 30 days, one row per day and model
curl "https://opendunes.com/v1/activity" \
  -H "Authorization: Bearer $OPENDUNES_API_KEY"

JSON

{
  "data": [
    {
      "date": "2026-06-30",
      "model": "anthropic/claude-opus-4.8",
      "model_permaslug": "anthropic/claude-opus-4.8",
      "endpoint_id": "a223cc6b-19a9-48ac-a297-3210a86073e5",
      "provider_name": "anthropic",
      "prompt_tokens": 482000,
      "completion_tokens": 123000,
      "requests": 1245,
      "usage": 184.23
    }
  ]
}

usage is the DA spent that day on that model. Pass ?date=YYYY-MM-DD to narrow to a single day.

Spend by model#

The rows arrive grouped per day × model, so a per-model breakdown is a small aggregation:

Python

import os, requests
from collections import defaultdict

BASE    = "https://opendunes.com"
HEADERS = {"Authorization": f"Bearer {os.environ['OPENDUNES_API_KEY']}"}

rows = requests.get(f"{BASE}/v1/activity", headers=HEADERS).json()["data"]

by_model = defaultdict(float)
for r in rows:
    by_model[r["model"]] += r["usage"]

for model, da in sorted(by_model.items(), key=lambda x: -x[1]):
    print(f"{model}: {da:.2f} DA")

This breakdown tells you which models are driving most of your spend — often the starting point for optimization.

Building a daily spend alert#

Python

def check_daily_spend(threshold_da: float) -> None:
    from datetime import date

    resp = requests.get(
        f"{BASE}/v1/activity",
        params={"date": date.today().isoformat()},
        headers=HEADERS,
    )
    resp.raise_for_status()

    total = sum(r["usage"] for r in resp.json()["data"])
    if total > threshold_da:
        send_alert(f"Spend spike: {total:.2f} DA so far today")

def send_alert(message: str) -> None:
    # Send to Slack, email, PagerDuty, etc.
    print(f"ALERT: {message}")

# Run this on a cron every hour
check_daily_spend(threshold_da=100.0)

Checking remaining balance#

Shell

curl https://opendunes.com/v1/credits \
  -H "Authorization: Bearer $OPENDUNES_API_KEY"

JSON

{
  "data": {
    "total_credits": 5000,
    "total_usage": 4575
  }
}

Automated low-balance protection#

Python

def balance_da() -> float:
    r = requests.get(f"{BASE}/v1/credits", headers=HEADERS)
    r.raise_for_status()
    data = r.json()["data"]
    return data["total_credits"] - data["total_usage"]

LOW_BALANCE_THRESHOLD_DA = 50.0

if balance_da() < LOW_BALANCE_THRESHOLD_DA:
    # Pause job queue, alert on-call, redirect to a cheaper model tier
    pause_inference_queue()
    send_alert(f"Balance below {LOW_BALANCE_THRESHOLD_DA} DA — topping up required.")

Cost optimization checklist#

Model downsizing — from the by-model breakdown, identify high-spend models and test whether a cheaper model meets your quality bar on those tasks.
Prompt compression — long system prompts multiply cost across every request. Trim where possible.
max_tokens caps — set a realistic max_tokens per request type; runaway generations can multiply expected cost by 5–10×.
Rate-limit tuning — if a batch job is generating thousands of requests in a short window, adding a delay reduces peak spend without affecting throughput.
Cheap models for triage — sort the catalog by price and use a low-cost model for classification and routing decisions that don't require a premium one.