> For the complete documentation index, see [llms.txt](https://docs.darcyiq.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.darcyiq.com/dispatch/cost-and-insights/insights.md). # Insights Checks on how your keys or your organization use Dispatch, what to do about each finding, and where Prompt Optimization is saving money Overview and Cost Reporting tell you what Dispatch is costing. **Insights tells you what's worth doing about it.** Dispatch runs a set of checks over the last 30 days of traffic, such as keys with no spending cap, keys nobody uses, spend with no tag, or agents resending large amounts of command output. It then lists what needs doing, most urgent first, with the next step for each and a button to the page that fixes it. Insights answers for whoever is reading it. A **Dispatch Admin** sees checks across the whole organization. A **Dispatch User** sees checks on the keys they own. {% hint style="success" %} **A to-do list that pays for itself**: Insights flags that your developers' agents are resending test runs and diffs on every turn, names the keys doing it, and estimates what that text cost to send. One switch in Guardrails turns on Prompt Optimization, and the coverage map on this page shows those keys' days filling in as tokens come off. {% endhint %} ## How Insights Differs by Role | | Dispatch User | Dispatch Admin | | -------------------------------- | ------------------------------------------------------------ | -------------------------------------------------------------------- | | Page description | "Checks on how your keys use Dispatch over the last 30 days" | "Checks on how the organization uses Dispatch over the last 30 days" | | Checks run | Checks about your own keys | Every check, including organization-wide ones | | **For the record** tab | — | ✓ | | **Prompt Optimization coverage** | Your keys | Every key, with an **Open guardrails** button | Organization-wide checks, such as the organization's spend limit or spend concentration, aren't run for a Dispatch User at all. They're about the organization rather than anyone's own keys, and only a Dispatch Admin can act on them. ## Running the Checks The checks run when you open the page. The header shows when they were last run, such as "Checked 3 minutes ago". Click **Run again** to rerun them, for example after changing a setting. ## The Summary Four tiles sit across the top. | Tile | What it shows | | ------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Needs attention** | Findings to act on. Click to show only these, and click again to show everything | | **Worth reviewing** | Findings to look at when you can. Click to filter the same way | | **Could save** | The estimated total across findings that carry a dollar figure, over the window. Only findings that could be priced are counted, and it shows a dash when nothing could be | | **Checks passed** | How many checks passed out of how many ran, and how many couldn't be checked yet | Below the tiles, category chips (**All**, then a chip per category, such as **Spending controls** or **Wasted spend**) narrow the list to one category. Each chip shows how many of that category's findings are on the to-do list. ## The To-Do List **To do** lists every finding that needs action, sorted by severity and then by potential saving, so the most urgent finding is at the top and opens automatically. Each row shows: * The finding, in a sentence, such as "Some keys are not being used" * Its category and severity, and how many people or keys are affected * The **Potential saving** on the right, where one could be priced Severity is always written out as well as shown in colour: **Needs attention**, **Worth reviewing** or **Informational**. Click a row to open it. {% stepper %} {% step %} **Read what was found** A short explanation of the finding, with the figures behind it. {% endstep %} {% step %} **Read what to do** **What to do** gives the next step. {% endstep %} {% step %} **See who's affected** The people or keys behind the finding, with a note on each. The first five are shown, and **Show all** lists the rest. Where a name is a link, click it to open that key or person on the API Keys page. {% endstep %} {% step %} **Act on it** Use the button to the page that fixes it, or **Copy for your developers** to copy the finding, the next step and the affected list as plain text for a chat or ticket. {% endstep %} {% endstepper %} Where the fix belongs in your developers' code, such as trimming long prompts, there's no page button, and **Copy for your developers** covers it. ### Passed and Not Yet Checked Under the list, **Passed** collects the checks that found nothing wrong. Some passes are also status reports, such as which keys a setting covers, and open like a finding so you can read the list. **Could not be checked yet** collects checks that couldn't run, each with the reason. These aren't passes. A check that couldn't read what it needs says "Nothing is implied either way". Often the remedy is time: some checks need a minimum amount of history before they can say anything. ## The Checks Each check has a short name, which is what you see when it passes or can't run. When it finds something, the finding is written as a sentence instead. ### Spending Controls | Check | Finds | Who gets it | Button | | ---------------------------------------- | ------------------------------------------------ | ----------- | ----------------------- | | **Spending caps** | Key owners who can spend without a limit | Everyone | **Open API keys** | | **Remaining allowance** | Key owners close to their spending cap | Everyone | **Open API keys** | | **Requests refused by a spending limit** | Requests turned away because a limit was reached | Everyone | **Open API keys** | | **Organization spend limit** | No overall spend limit for the organization | Admins | **Open budget** | | **Budget pace** | Budgets on course to run out at the current rate | Admins | **Open budget** | | **Unusual days** | Recent days that cost far more than usual | Admins | **Open Cost Reporting** | ### Key Security | Check | Finds | Who gets it | Button | | ------------------------- | ---------------------------------------------------------- | ----------- | ----------------- | | **Keys returning to use** | A key that had gone quiet for a long time being used again | Everyone | **Open API keys** | ### Wasted Spend | Check | Finds | Who gets it | Button | | -------------------------------------- | --------------------------------------------------------------------- | ----------- | ------------------- | | **Repeated prompt text** | Paying to send the same text over and over | Admins | — | | **Tier choice** | Traffic that may be running on a dearer tier than it needs | Admins | **Open Models** | | **Requests for tiers you do not have** | Requests turned away for naming a tier your organization doesn't have | Admins | **Open Models** | | **Rejected requests** | Requests that couldn't be read at all | Admins | — | | **Abandoned answers** | Answers cut off before they finished | Admins | — | | **Prompt size against answer size** | Very long prompts for very short answers | Admins | — | | **Command output** | Agents sending a lot of command output, such as test runs and diffs | Everyone | **Open guardrails** | | **Bulk data from tools** | Tools returning large lists of records | Everyone | **Open guardrails** | | **Tool list size** | Every request carrying a long list of tool descriptions | Everyone | — | The last three only appear when Dispatch has measured requests that use tools. ### Unused Keys | Check | Finds | Who gets it | Button | | ------------------------------- | ------------------------------------------------------------ | ----------- | ----------------------- | | **Key owners who have stopped** | Key owners with working keys who have stopped using Dispatch | Everyone | **Open API keys** | | **Unused keys** | Keys that made no requests in the window | Everyone | **Open API keys** | | **Keys per owner** | Key owners holding an unusual number of keys | Everyone | **Open API keys** | | **Spend concentration** | Most of the organization's spend going through a single key | Admins | **Open Cost Reporting** | ### Compliance | Check | Finds | Who gets it | Button | | ---------------------------------------- | -------------------------------------- | ----------- | ------------------- | | **Content policy coverage** | Keys that run without a content policy | Everyone | **Open guardrails** | | **Requests blocked by a content policy** | Requests a content policy refused | Everyone | **Open guardrails** | ### Cost Allocation | Check | Finds | Who gets it | Button | | ---------------- | ------------------------- | ----------- | ----------------- | | **Tag coverage** | Spend on keys with no tag | Everyone | **Open API keys** | **Open budget** and **Open guardrails** go to [Billing & Budget](/dispatch/manage/manage/billing-and-budget.md) and [Guardrails](/dispatch/manage/manage/guardrails.md). **Open Cost Reporting** opens the **Spend by person, by week** report. See [Cost Reporting](/dispatch/cost-and-insights/cost-reporting.md). ## Prompt Optimization Prompt Optimization shortens what your team's agents and apps send to the model, before it's sent. It covers two kinds of content: * **Command output** that can be shortened without losing anything, such as a test run where every test passed, a diff, a search or a file listing * **JSON tool results**, re-sent in a compact layout that keeps every value Failures and errors are always sent in full, and file contents are never changed. Anything Dispatch can't shorten safely goes through exactly as it arrived. ### Turning It On Prompt Optimization is off until a Dispatch Admin turns it on. The organization's setting is the **On for this organization** switch on the **Prompt Optimization** card in Organization › Guardrails. Each guardrail can then **Inherit organization setting**, or be set to **On** or **Off** for the people under it. A change applies within a minute. See [Guardrails](/dispatch/manage/manage/guardrails.md). The **Command output** and **Bulk data from tools** checks suggest turning it on when your keys send a lot of either. Once Prompt Optimization is on anywhere in the organization, the **Command output** check becomes a status report instead, saying "Prompt Optimization is on for … of … keys" and listing what each key saved. ### How the Saving Is Measured Each request's saving is priced as it happens, at what the removed tokens would have cost on the model that answered it. Where part of the prompt was served from cache, that part is priced at the cheaper cached rate. You'll find the total under **Total savings** and **What Dispatch saved** on [Overview](/dispatch/use-dispatch/overview.md), and as **Saved by Prompt Optimization** in [Cost Reporting](/dispatch/cost-and-insights/cost-reporting.md). ### Prompt Optimization Coverage Below the to-do list, **Prompt Optimization coverage** shows which keys it reaches and what it did, day by day. It appears once there's traffic to show. The header sums it up: whether it's **On for every key**, **Off for every key** or **On for** some of them, and once anything has been shortened, how many tokens were removed over the last 30 days and what they were worth. Each row is one key, labelled with its owner and the key's name, and the tool that uses it where Dispatch recognizes it (for example **Coding agent** or **App**). Each square is one day: | Square | Meaning | | ------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Coloured | Tokens were removed that day. The colour shows what was shortened (**Command output compression**, **Tool result compression** or **Other compression**), and darker means more tokens removed | | Grey | On for this key, but nothing was shortened that day | | Outline only | **Off for this key** | Hover a square for that day's figures. On the right of each row: * **Keys it's on for** show the tokens removed and what they were worth, or **Nothing shortened yet**, with why it's on: **On**, or **On via guardrail '…'** * **Keys it's off for** show why, such as **Off: organization default is off** or **Off: guardrail '…' turns it off**, and how much command output and tool results that key sent which Prompt Optimization could have shortened Keys it's on for come first, ordered by what they saved, then the rest by what they could have saved. Ten rows show at first, and **Show all** reveals up to 40. Any keys beyond that are counted as "quieter keys not drawn". Dispatch Admins can click **Open guardrails** to change the setting. ## For the Record (Dispatch Admins) The **For the record** tab isn't a list of things to fix. It's evidence you can forward when someone asks what your AI vendor stores or what your content policy covers. | Panel | What it shows | | --------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | **What we store** | That Dispatch doesn't store the text of prompts or answers, doesn't store client IP addresses or anything derived from them, and keeps API keys only as a one-way digest that can't be read back | | **Content policy coverage** | How many key owners send their requests under a content policy, each policy that's in use and what it checks for, and that matching requests are refused inside Dispatch before they're sent to any model provider | | **Where requests ran** | A note that Dispatch doesn't record the region each request was processed in, so it doesn't report one | To control where requests may be processed, see [Data Residency](/dispatch/manage/manage/data-residency.md). ## About the Dollar Figures Every dollar figure on Insights is in the same terms as **Total cost** on Overview and **Spend** in Cost Reporting. **Could save** and each finding's **Potential saving** are estimates over the window, and for some findings, such as **Command output**, the figure is what that content cost to send. ## Forecasts, Savings and Habits Some related views live on other pages: | View | Where | | ------------------------------------------------------------------ | --------------------------------------------------------------------- | | **The next 30 days**, the spend forecast | [Overview](/dispatch/use-dispatch/overview.md#where-spend-is-heading) | | **What Dispatch saved**, day by day | [Overview](/dispatch/use-dispatch/overview.md#what-dispatch-saved) | | **Your habits** and **Habits across the organization** | [Overview](/dispatch/use-dispatch/overview.md#habits) | | **Intelligence vs. Price**, comparing tiers with well-known models | [Models](/dispatch/use-dispatch/models.md) | ## Next Steps | Goal | Documentation | | --------------------------------------------------- | ----------------------------------------------------------------- | | Turn on Prompt Optimization or add a content policy | [Guardrails](/dispatch/manage/manage/guardrails.md) | | Set spending caps and fix unused keys | [API Keys](/dispatch/use-dispatch/api-keys.md) | | Set the organization's spend limit | [Billing & Budget](/dispatch/manage/manage/billing-and-budget.md) | | Tag keys so spend can be allocated | [Tags](/dispatch/manage/manage/tags.md) | | Dig into a finding with your own breakdown | [Cost Reporting](/dispatch/cost-and-insights/cost-reporting.md) | | See the forecast and savings over time | [Overview](/dispatch/use-dispatch/overview.md) | --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.darcyiq.com/dispatch/cost-and-insights/insights.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.