Data Dashboard
Total execution volume, success rate, average time consumption, token usage and Top workflow - account-level operating data can be viewed on one page, and single execution cost estimates can be queried.
The console's Run Analytics page rolls all runs in a time window into one dashboard: four KPI cards, two line charts for run and duration trends, a donut chart of run status distribution, and a top-workflows table sorted by run count. A morning glance answers "how many ran yesterday, did they succeed, and where did the tokens go".
Opening the Dashboard and Time Ranges
Find it in the console sidebar → Run Analytics. The top-right corner switches between three time ranges: 24 hours (aggregated hourly), 7 days, and 30 days (both aggregated daily). Counting rules: a run is included if its start time falls inside the window, and only runs the current account is allowed to see are counted.
Dashboard data has a short browser-side cache of about 30 seconds; switching time ranges renders instantly on a cache hit, and the refresh button in the top-right corner bypasses the cache to pull fresh data.
The Four KPI Cards
| Indicator | Definition |
|---|---|
| Executions | Number of runs in the window — running, failed, and canceled all count |
| Success Rate | Runs with status COMPLETED ÷ total runs |
| Average Time Taken | Average duration of finished runs that have a recorded duration |
| Total Token Usage | Sum of tokens recorded in step events across all runs in the window (input + output); K is thousands, M is millions |
Every turn of an agent step re-reads the same prefix, which hits the model's prompt cache. Total token usage counts those cached reads too — so a multi-step agent run is usually far larger than "what you typed in". That is what the provider actually processed.
Trend Charts and Status Distribution
- Execution trend — a line of execution counts per time bucket (hour or day), for reading business cadence and sudden spikes
- Average duration trend — the mean execution time (seconds) per time bucket; a slowly rising curve usually means an upstream API got slower or steps were added to the workflow
- Execution status distribution — a donut chart of succeeded / failed / cancelled, for overall health
The donut chart only shows executions that have been classified as successful, failed, and canceled; executions that are still running, waiting for approval, or in an interrupted state will be included in the "Total Executions" and the success rate denominator, but will not appear in these three sectors, so the total number of sectors may be less than the total number of executions.
Top Workflow
The table at the bottom of the dashboard lists workflows by execution count, high to low. Each row shows execution count, success rate (with a colored progress bar), average duration, last run time, and token usage. To find where tokens are going or which workflow's success rate has dropped, start here.
How Costs Are Calculated
Each execution's cost is an estimate, computed from the token counts actually recorded in execution events:
成本 ≈ 输入 token × 输入单价 + 输出 token × 输出单价
（单价 = 美元 / 100 万 token，按模型查内置单价表）- Unit prices come from the platform's built-in price table, in USD per 1 million tokens, with input and output priced separately
- The price table covers common models from Anthropic, OpenAI, Google, DeepSeek, Qwen, Mistral, and others, matching model names fuzzily
- The endpoint lists input and output tokens plus the estimated cost for each step, making high-cost steps easier to identify.
# 单次执行的成本估算（含逐步骤拆分）
curl "https://braidrun.com/api/executions/<execution-id>/cost" \
-H "Authorization: Bearer <token>"{
"executionId": "<execution-id>",
"totalCostUsd": 0.0837,
"modelName": "anthropic/claude-3.5-sonnet",
"inputTokens": 12400,
"outputTokens": 3100,
"totalTokens": 15500,
"stepCosts": [
{"stepName": "fetch_data", "costUsd": 0.0, "inputTokens": 0, "outputTokens": 0},
{"stepName": "daily_summary", "costUsd": 0.0837, "inputTokens": 12400, "outputTokens": 3100}
]
}- The price table is built into the platform and may lag behind provider price changes; models not in the table are counted as 0
- The current per-execution cost endpoint identifies one model from execution events and uses that model’s unit price to estimate every step. It falls back to the default model when no model can be identified. Treat the result as a rough estimate for workflows that use multiple models.
- Under BYOK, actual charges land in your own provider account, and your provider's bill is the source of truth for reconciliation
- The platform does not quote costs before a run — cost is only computed after execution, from the actual token records
Pull the Same Data via the API
The aggregated data behind the dashboard can also be requested directly, so you can wire it into your own reports or alerts:
# 看板同款聚合数据（KPI + 趋势 + 逐工作流统计）
curl "https://braidrun.com/api/dashboard/analytics?range=7d" \
-H "Authorization: Bearer <token>"
# Top 工作流（limit 默认 10，最大 100）
curl "https://braidrun.com/api/dashboard/top-workflows?range=30d&limit=10" \
-H "Authorization: Bearer <token>"The range parameter is a number plus a unit, such as 24h, 7d, or 30d. The analytics response contains a KPI summary, two time series (executions and duration), and per-workflow statistics (perWorkflowStats).
Related Pages
- Execution Monitoring and Logging — timeline, logs, and token details for a single execution
- BYOK: Bring Your Own Key — token costs are billed to your own provider account