Tally evidence attachments
Add images, PDFs, CSV/XLSX spreadsheets, and existing LedgerHQ files so Tally can inspect evidence safely.
Tally can use evidence attached to the full-page chat. Evidence can be a pasted screenshot, an uploaded image, PDF, CSV, or XLSX file, or an existing file selected from LedgerHQ sources such as Tally uploads, Documents, accounting attachments, and stored bank statements.
Use evidence when the question depends on something visual or source-backed: a screenshot of a confusing row, a statement PDF, a client upload, a report export, or a document that explains how a transaction should be treated.

Add New Evidence
In the full-page Tally workspace, add evidence by pasting a screenshot, dragging a supported file onto the composer, or browsing for a supported file. The composer can send text-only, attachment-only, or text-plus-attachment messages.
When an upload is still in progress, wait for the attachment card to finish. LedgerHQ uploads the file through the app server and stores it before creating the Tally evidence record. That keeps browser-to-storage failures from leaving the user stuck with an unclear pending attachment.
Search Existing Evidence
Use evidence search when the file already exists in LedgerHQ. Tally can search across current Tally evidence, the Documents library, accounting attachments, and stored bank statements. Selecting an existing file links it to the Tally message without duplicating the underlying stored object.
This is useful when a statement, receipt, or prior upload already exists and the user wants Tally to inspect it in the current conversation.
What Tally Can Read
Current-message images and PDFs are sent to the model as active evidence. Immediately prior evidence may also be treated as active when the user's next message clearly refers to it, such as "Can you read that?" or "What does this say?"
Older evidence remains available as compact context and can be reselected when needed. That keeps the live conversation focused while preserving the audit trail.
CSV and XLSX files use a different path. Tally saves them into the firm's durable workspace, reads rows through the bounded spreadsheet tools, and can use a protected sandbox for calculations or transformation. The raw workbook is not pasted wholesale into the model conversation. Files Tally creates in that workspace can be reused by later turns until they are deleted or expire under the storage policy.
Evidence Is Not Authority
An attached file is evidence, not an instruction source. If a PDF, screenshot, or document contains text that says "ignore prior rules" or "post everything", Tally should treat that text as document content, not as a command.
Book-changing actions still require the normal guard paths: duty mode, company scope, authority, approval, period locks, posting validation, and audit records.
Do not upload secrets or credentials to Tally chat. Bank statements, receipts, and bookkeeping documents are expected evidence; passwords, API keys, Plaid credentials, Stripe secrets, and raw tokens are not.
Good Evidence Messages
A useful evidence message names what the user wants Tally to inspect. For example:
- "Read this statement and tell me whether the ending balance matches the rec."
- "This screenshot shows the row I think posted to the wrong account."
- "Use this PDF as support for the missing statement request."
- "Compare this export to the report I just generated."
- "Analyze this CSV, group the vendors, and show me the unresolved patterns before you change any books."
If Tally cannot read the file or the file does not match the expected company, account, or period, create a support ticket with the uploaded evidence and the specific mismatch.