AI-native reporting
Scan results written up for finance and engineering
Claude writes a CFO and an SRE narrative for each scan on a paid plan, a morning brief built on your remediation plan, and boards from a plain-English description.
It does not email the narratives or post them to Slack.
The problem
A dashboard shows what changed, not what it means. Someone still has to turn findings into a story for finance and a work list for engineering, every week.
How it works
Scan
A scan collects cost and usage, runs the rules and builds a remediation plan. That is the source material every narrative is written from.
Write
Claude writes the CFO and SRE narratives from the findings and plan, with identifiers such as ARNs swapped for placeholders first.
Read
Reports appear on the Reports page. The brief appears on your dashboard, and a notification tells you when it is ready.
2 narratives
written by Claude on a paid plan, from a scan’s findings and remediation plan
08:00
in your workspace’s time zone, when Claude writes a short brief
9 widgets
in the fixed list Claude picks from, and the server checks the layout
Up to 4
attempts by default, within 90 seconds, for overloads and rate limits
Capabilities
CFO and SRE narratives on paid plans, a morning brief, and boards you describe in words.
Two narratives per scan
On a paid plan, each completed scan gets a CFO narrative on dollar impact and risk, and an SRE narrative by service with remediation steps. With two or more cloud accounts, each account that has findings gets its own pair too.
A morning brief
At 08:00 in your workspace’s time zone, Claude writes a short brief: the most urgent item first, the savings in your remediation plan, pending approvals and high-severity findings, with links to each.
Boards from a description
Describe the board you want. Claude picks widgets from a fixed list of nine, the server checks the layout, and you save it. A layout that fails the check is rejected, not saved.
Routing and retries
Each prompt is matched to one of three tiers of Claude models. Overloads and rate limits are retried with backoff, and a narrative that still fails is saved as a clearly labelled template instead.
In depth
What the report narratives cover
Every completed scan ends with a report. On a paid plan, Claude writes two narratives for it from that scan’s findings and remediation plan.
- The CFO narrative: three to five paragraphs on dollar impact and risk, using savings by category, verified savings, commitment coverage, budget status and AI spend where that data exists. It is told to leave out technical identifiers.
- The SRE narrative: findings grouped by service, with metrics, remediation steps and AI spend per model. Low-confidence findings are flagged, so the narrative suggests more monitoring before acting on them.
- With two or more cloud accounts, each account that has findings gets its own report beside the workspace report. Accounts skipped for having no findings are recorded in the audit trail.
More ways to use a report
On the Reports page, a finished report can be taken further. Each of these calls Claude and needs a paid plan or an active trial.
You can also switch any report to a redacted view before sharing it.
- Regenerate the narratives.
- Write a narrative for another reader: a CISO, DevOps lead, board, product manager, engineering VP, or one you name.
- Compare two reports to see new and resolved findings.
- Get a five-point executive brief to paste into Slack or email.
- Ask questions about the report in a chat panel.
- Write a narrative across all your cloud accounts from the cross-account view.
The daily brief
The brief is a short note for an engineering lead: a greeting, three bullets and one next step. A scheduler checks every 15 minutes and writes it when your workspace’s clock reads 08:00. Paid workspaces, including trials, get it every day. Free workspaces get it on Mondays.
Its savings figure comes from the remediation plan: actions that can run automatically plus actions that need approval, never disallowed ones. It is not a total of every finding, because findings can price the same spend twice. A change is described only when two plans can be compared.
On a paid plan, a workspace with two or more cloud accounts also gets a brief for each account that has findings, written to say which account it covers. If the selected account has no brief, the dashboard shows the workspace brief and says so. The Brief page keeps earlier days and has a Regenerate button.
Boards you describe
Type what you want to see, such as “idle RDS by monthly savings, plus pending approvals”. Claude returns a layout built from nine widget types: savings, findings and cost-trend KPIs, a cost-trend chart, a findings table, severity and category charts, an approvals table and a sustainability summary.
The server checks every widget and filter against that list, up to 12 widgets, and a layout that fails is rejected. For this step Claude receives only your description and the widget list, not your cost data.
Each board has an “Ask this board” chat that answers from that board’s filters and your latest findings, report and approvals.
How the model is chosen
A router scores each prompt on its length, wording, structure and use of tools, and maps it to one of three tiers of Claude models. Some features are fixed to a tier: report narratives use the middle tier, and audience narratives and report comparisons use the top one.
The narratives, the brief, the board builder and the chats share one retry policy. Overloads, rate limits, server errors and network failures are retried with backoff and can move up one tier. A retired model is retried at once on the default model. Bad requests and authentication errors are not retried.
What it does not do
- It does not write Claude narratives on the free plan. Free workspaces get a template summary built from the findings instead.
- It does not email the narratives or post them to Slack. On the hosted service, a scan that collected data sends a summary instead: the workspace ID, the scan’s run ID and date, finding, severity and action counts, estimated monthly savings, the accounts not fully collected (a Slack message lists up to 20) and a link to the Reports page. It is emailed to your notification addresses and posted to a Slack incoming webhook if one is saved. Report emails are limited to about 10 a day. A self-hosted install emails it only when a mail sender is configured, and links to the Reports page only when its web address is configured.
- It does not guarantee every sentence is right. Claude is told to use only the numbers it is given, and the findings and plan beside each narrative remain the record.
- It does not swap resource names, tags or labels for placeholders. An instance named after a person is sent under that name.
FAQ
Questions
Which plan do I need?
Claude-written report narratives, regenerating them, audience narratives, report comparisons and report chat need a paid plan, and an active trial counts. Free workspaces get template reports, a weekly brief on Mondays and the board builder.
What data does Claude see for the CFO and SRE narratives?
The findings and remediation plan for that scan and, where that data exists, budget names with their spend against each limit, verified savings totals, commitment coverage and recommendations, and AI spend by model. In the findings, the plan and budget names, ARNs, AWS account numbers, common EC2 resource IDs, Google Cloud resource paths and service-account addresses, your workspace ID and each connected account’s number or project ID are swapped for placeholders before sending. The swap is pattern-based: resource names, tags and labels are sent as they are. A narrative for one cloud account gets that account’s budgets, with workspace-wide budgets listed separately.
What happens if Claude is unavailable?
Overloads, rate limits, server errors and network failures are retried, by default up to four attempts within 90 seconds. If the narrative still fails, the report is saved with a template summary marked as a fallback. The findings and plan are unaffected.
Can I get a narrative for one cloud account?
Yes, in a workspace with two or more accounts. Each scan writes a report for every account that has findings. On a paid plan, each of those accounts also gets its own brief, and the brief follows the account you select in the header.
Which model writes the reports?
Claude. A router matches each prompt to one of three tiers of Claude models. Report narratives use the middle tier, and a retry during an overload can move up one tier.
Try free — no credit card
Create a workspace and connect an AWS or GCP account, or book a demo first.