Ask GageLog

Ask GageLog is a chat assistant built into GageLog. Ask about any public gage in plain English and it reads the same 15-minute data you see on the site, runs the analytics for you, and can start reports or manage your alerts. It is included with the Field plan and above. The overview page has examples; this page covers how it works.

Opening it

• The Ask GageLog tab. On the tracked gages pages, open the Ask GageLog tab for the full chat with your conversation list on the left: start a new conversation, rename, delete, or export.
• From a gage page. The Ask button on a gage's detail page opens a panel already pointed at that gage, so “how is it compared with last year?” needs no gage name.
• From Home. The quick-ask box on Home is the fastest way to check on your gages. Suggested starter prompts are built from the gages you track.

On Free and Riverside the chat shows a preview and a link to upgrade.

Prompting tips

• Name the gage and the window. “Dungeness near Sequim, last 7 days” beats “how's the river?”. Gage names, keys or station numbers all work.
• Ask for a comparison, not two answers. “Compare the Skykomish and the Snoqualmie since Friday” returns them on one footing.
• Give a threshold. “How many days this summer was the Arkansas at Salida under 500 CFS?” gets an exact count rather than a vague answer.
• Say what you want back. A short summary, a table, or a report you can download.
• Start a fresh conversation for a new topic. Long threads are capped, and a focused conversation answers better.

What it can read

Ask GageLog uses the same read tools as the MCP server, across every public gage in the registry. In everyday terms:

• Finding gages — your tracked gages, and the gages available to you.
• Readings — the latest reading, any date range of 15-minute readings, and water-year daily values and statistics.
• Summaries and trends — flow summaries, rolling averages, rising and falling trends, and data-quality checks.
• Thresholds and volumes — how often a gage was above or below a level, and how much water passed between two dates.
• Comparisons — two or more gages side by side, or a multi-gage summary.
• Dry Year Leasing periods — the monitoring-window calculations used for the Dungeness DYL report.
• Gage details and your plan — metadata about a gage, plus your plan and usage.

What it can do for you

Actions only ever apply to your own account, and they follow the same rules as the rest of the site: your plan's tracked-gage and alert caps, slot cooldowns and swap credits, and export availability by plan.

• Track a gage — done directly if you have an open slot.
• Untrack a gage — asks for confirmation first.
• Create or edit a flow alert — thresholds, anomaly alerts and offline-gage alerts, delivered the way the Monitors page delivers them.
• Delete an alert — asks for confirmation first.
• Start a report or export — see below.

Reports and exports

Ask in plain words and it picks the matching export: the one-page gage report, the flow-conditions PDF, a CSV or Excel workbook of readings, a multi-gage comparison, statistics or data workbooks, a visualization workbook of your tracked gages, or one of the Flow Workbench templates. Which of these your plan includes is the same as on the Exports page.

Reports are built in the background. The chat shows a report card with its status, a Download button when it is ready, and the file also appears in your Exports. You can keep chatting while it builds. Up to three exports build at a time; ask for more and the rest wait in line and start automatically as earlier ones finish — their cards read “Waiting for a free slot” until then. You never need to ask again. For a report over many gages (for example the flow-conditions PDF for all your gages, which covers up to 8 per file) it queues one report per batch in a single reply.

Confirmations

Anything that cannot be undone, or that costs something, stops and shows a card with Confirm and Cancel buttons: untracking a gage, deleting an alert, and any change that would spend a swap credit. Nothing happens until you click Confirm. A card that you leave for a while expires, and you can ask again. Your choice is passed back to the assistant so it knows what happened on the next turn.

Limits and counters

• The meter shows what is left today, for example “37 of 100 left today”, and when it resets. The daily count resets at midnight in your time zone.
• The monthly token meter counts what the model reads and writes. It resets at the start of the month.
• A prompt is counted when you send it. If the failure is on our side or the model service's, it is given back. A prompt that is declined by the safety screen, or that the model refuses to answer, still counts.
• Conversations are capped at a set number of turns. When you reach it, start a new conversation.
• Team admins see how many prompts and tokens each member used, never what they asked.

Managing and exporting conversations

Rename a conversation from the sidebar, delete it, or export it as Markdown or PDF. Conversations are deleted automatically 90 days after their last message (a rename does not extend that). A thumbs up or down on any answer helps us tune it.

Limitations

• It is an AI assistant and can be wrong. Check any number you will rely on against the gage page, which is why every figure is quoted with its gage and timestamp.
• It does not make safety calls. It will not tell you whether a river is safe to boat, wade or cross. Scout it and use your own judgment.
• It stays on topic: GageLog, rivers, streamflow and hydrology, and your gages, alerts, reports and plan. Other requests get a short refusal.
• It cannot see other users' data and does not discuss how GageLog is built.
• It can only act on your account, within your plan's limits.

Data accuracy

The answers come from the same readings as the rest of GageLog. Recent readings are provisional: agencies review and revise them, sometimes days later, and a gage can be offline or reporting stale values. Ask GageLog notes provisional data and flags gaps it sees. See Data sources & methodology. Do not use it for regulatory filings or binding water-rights decisions without verifying the data at the source.

Troubleshooting