Let an AI assistant answer questions using selected emails, documents and data. Create reusable groups for supplier reviews, contract questions or other work, choose what each connection can read and see a history of its requests. Connect compatible tools through MCP or REST API.
Cedar Works · Delivery update · 10 October
Delivery is planned for 18 October 2026.
Delivery contact: Mira Cole
Agreement: Cedar Works supply contract
Internal price: £12,400
The assistant can read the delivery details while internal prices and original documents stay private.
Context for the work your team does
Create a custom list of supplier emails across your inboxes, gather the files for a contract review or save the documents behind a customer question. Connect the accounts and folders where that information lives and reuse the group when another question comes up.
Share selected emails and files, a table of extracted data or the current details for a supplier or contract. Give each team or AI tool its own view, so you can reuse the information without assembling a new export for every question.
Clear fields and supporting documents
Tell the connected tool what each row represents and what its fields mean, including dates, currencies and units. A supplier count and an invoice-line count can then refer to the right things. These definitions give the assistant a clearer basis for its questions and calculations.
Include references to the supplier email behind a delivery date or the contract behind a notice deadline. Where access permits, the assistant can return those references so a person can check the answer against its source.
Choose accepted rows for a reviewed-data view. When a person must check important amounts or deadlines, use recorded field reviews as well: your extraction setup can also accept results automatically.
Keep sensitive details private
Choose the fields to share and remove, mask or replace sensitive values. Use consistent aliases when a tool needs to relate items without receiving real identifiers. The original information stays unchanged, and your sharing policies apply to MCP and REST searches, results and calculations.
Choose separately whether the assistant can receive supporting references, open source documents or download original files. Share a delivery date while keeping prices private. Hidden raw values are excluded from search and filter decisions; masking supported fields or text leaves original PDFs and attachments unchanged.
Questions about numbers and changes
Ask for totals by supplier, average invoice amounts or counts by month. Polytrace calculates supported results from the structured fields before sending them to the AI tool, using the data after your sharing policies have been applied.
Use supported formulas for ratios, date differences and other calculations, with checks for field types, units and currencies. Fixed snapshots support exact calculations. Live information includes a change warning, and requests beyond the processing limits return an error rather than an incomplete total.
Show whether the assistant is reading the latest captured information, a scheduled update or a saved snapshot. Scheduled and fixed views use their completed versions so a question about last week’s information stays tied to that version.
Compare retained versions to see which items were added or removed and which fields changed. Search current items by their designated date when you need recent information, and use retained versions to compare earlier results.
Connect tools and review their access
Connect compatible AI assistants and applications through read-only MCP or REST API access. Let them search emails and files, retrieve selected fields, calculate results and cite supporting documents while keeping account credentials private.
Give a connection access to one shared view, or let it discover the eligible views in your organization’s context catalog. Set how much it can retrieve or calculate and revoke its credentials when access is no longer needed.
Browse, search and count the shared items in the Context playground. Inspect the returned fields and copy a JSON example to see how the connection will present your information, without calling a model.
Review the intended client’s access, including masked values, original files and supporting references. The AI context guide covers connection setup. You can serve context to an external assistant without connecting an AI provider inside Polytrace.
Review who requested information, when they requested it, the tool used, the stated purpose, the number of results and whether the request succeeded. See which sharing policies were applied. You can require a purpose for requests that support one; standard MCP search and fetch use their recorded server-defined purposes.
Trace requests to the selected data and saved version through audit references, without storing full requests or responses in the Audit Log. For the external AI provider’s later use of that information, review its own terms and your agreement.
You can share selected emails and files or fields extracted with fixed rules. Use AI extraction when the wording needs interpretation. Serving information to an external assistant does not itself call a model inside Polytrace.
Use a compatible MCP or REST client with the supported credentials. MCP clients need bearer-token support; clients that require OAuth need a separately deployed OAuth connection. The connection guide helps you check your intended client.
These connections have read-only access to the information you share. They can search and retrieve it, with no permission to edit your emails, files or connected sources. Returned content is identified as data for the client.
Masking changes supported shared fields and text while leaving original files intact. Original downloads are enabled by default for eligible file and attachment views, so turn them off when a connection should receive only the selected information.
Use clear field definitions, calculations performed by Polytrace and permitted references to the original emails or documents. These give you ways to check an answer; the model’s interpretation still needs review for important decisions.
Polytrace controls the data it returns. The connected provider’s retention, training and answer behavior depend on that provider and your agreement with it. Review those terms when choosing the connection.
Try a supplier, contract or invoice question your team answers today. See the assistant’s permitted information, the source behind its answer and the request recorded in the access history.
