Skip to content

AI Assistant

Eclipse AI is designed to work like an operator, not just a chatbot. You can ask a question in plain English, upload supporting files, and let Eclipse build an editable workbench for the next step.

Typical examples include:

  • finding sales that still do not have COGS linked
  • spotting possible deductions that are missing from expenses
  • reconciling 1099 totals against what Eclipse already has recorded
  • turning uploads into draft expenses or cleanup queues
  • reviewing unmatched sales, products, or import rows in one table

The AI workspace also includes an Eclipse service rail. It shows work Eclipse has found from real accounting data, summarizes what the service completed, and keeps anything outside your guardrails visible for review.

  1. Open AI from the app and ask a focused business question 1. Upload the files that matter for the task, such as PDFs, CSVs, XLSX/XLS workbooks, or a ZIP bundle 1. Choose whether you want Eclipse to only analyze, auto-apply low-risk edits, or auto-apply low-risk edits and create drafts 1. Review the workbench Eclipse opens alongside the conversation

The workbench behaves like a focused spreadsheet for the question you asked. Depending on the task, Eclipse can open workbenches such as:

  • Missing COGS
  • Possible deductions
  • 1099 reconciliation
  • Draft expenses from uploads
  • Unmatched sales and products

Each row includes the working data Eclipse found, the current status, and the source behind the suggestion so you can review the result without leaving the chat flow.

Tax deduction opportunities also show confidence, source transactions, the exact status action, ledger guardrails, recovery guidance, and audit posture before you claim or dismiss an insight. Eclipse records shown and edited deduction-suggestion events without storing raw suggestion text.

The Audit > AI History tab shows recent shown, edited, accepted, rejected, generated, and updated AI events from banking, invoices, expenses, reports, and tax without exposing raw suggestion payloads. Report explanation prompts include cited source rows when you use Ask about this report from the Reports workspace. Customer follow-up drafts launched from customer profiles and quick views also include cited customer rows for balances, health, activity, and open tasks.

Eclipse treats uploads as working inputs, not just reference files.

  • PDFs and images can be read for text and tax details
  • CSV and spreadsheet uploads can become previewable tabular datasets
  • ZIP bundles are expanded file by file, with unsupported or failed files shown clearly in the dataset summary
  • tax-related uploads such as 1099s and payment confirmations get first-pass extraction for totals, years, and key references

The assistant separates actions by risk so it can move quickly without making unsafe accounting changes.

  • Analyze keeps the session read-only and builds the workbench without applying changes.
  • Auto-apply low risk lets Eclipse handle safe cleanup work such as categorization, linking, normalization, and other non-posting edits.
  • Auto-apply and create drafts does the same low-risk cleanup and can also create silent draft records like expenses, import rows, or reconciliation candidates.

High-risk actions do not run automatically. Anything destructive or financially final stays in a review-required state instead of posting itself.

On wider screens, the AI workspace shows two service cards beside chat:

  • Eclipse service shows whether the service is on, the current autonomy mode, recent work completed, open items that need you, and guardrail blocks.
  • Work Eclipse found lists live org-scoped work such as missing COGS, pending approvals, and expense-review items.

The service can run in watch, draft, auto-safe, or full-auto mode. Auto modes still use amount and confidence guardrails, and posting-sensitive work goes through the normal approval and ledger paths.

For month-end close, Eclipse will not call a period ready when its checklist is missing, required work remains, or ledger integrity checks fail. Full-auto can soft-close only after those controls pass; a failed workspace run is reported as a failed scheduled run instead of a successful empty result.

Every operator run keeps an activity log with the current batch summary.

  • auto-applied edits are counted and can be undone from the batch log
  • draft records stay visible in the same activity summary
  • review-required actions stay pending until a person approves the next step
  • the workbench remains available so you can reopen the same session later

The AI assistant stays inside the currently authorized workspace scope. It uses Eclipse data first when available, then combines uploads only within that same workspace context.

Eclipse can help investigate, prepare, and draft work, but it does not silently post journal entries, delete records, or finalize tax outcomes for you.

Eclipse uses two complementary AI layers:

  • Platform AI — the chat rail and /ai page for investigation, charts, cited answers, budget variance explanations, and batch cleanup work
  • Quick Assist — inline sparkle actions on individual fields and rows, such as generating an invoice description or suggesting a category on one bank transaction

Use Ask Eclipse from reports, customers, budgets, and other major surfaces when you want Platform AI with the current page context and cited source rows already attached.

Eclipse AI receives structured context on every chat turn:

  • Business profile — your company name, industry, accounting basis, fiscal calendar, and connected integrations
  • Current screen — active filters, summary counts, and cited rows from the page you are on (dashboard, banking, invoices, reports, tax, approvals, and more)
  • Detail records — when you open an invoice, expense, or other detail page, Eclipse loads a short org-scoped summary before answering

Financial numbers still come from live data tools; screen context tells Eclipse what you are looking at so you do not have to paste IDs or re-describe filters.

When you ask for a chart, trend, or breakdown, Eclipse can return an inline chart or table in chat, with source rows you can open from the answer. Saved reports keep the chart configuration when you open them in the report builder.