These silent recordings use synthetic mail. The notes describe visible steps so you can follow the terminal text without relying on the video. Agent replies are excerpts from a recorded session; they are not a test of current model quality. The TUI clips show older layouts, and the edited montages combine separate recordings.
Your browser cannot play this video. Open the recording .
Ask an agent to inspect suspicious mail 43s A recorded agent session searches a synthetic inbox, checks message headers, and reports suspicious senders.
Excerpt from a recorded synthetic-inbox session. GPT-5.5 is visible in the capture; the recording date and mxr version are not identified.
Read the recording notes The prompt asks for likely phishing or scam senders and says not to change anything. The agent searches inbox messages for suspicious phrases and inspects selected headers. It returns a short list of senders to verify; the clip does not establish current model quality. Open video: Ask an agent to inspect suspicious mail Your browser cannot play this video. Open the recording .
Have an agent search the synthetic inbox 20s An excerpt shows an agent form a mail search and inspect matching messages from a synthetic inbox.
Excerpt from the same recorded synthetic-inbox agent session as clips 01, 03, and 04. GPT-5.5 is visible; date and mxr version are not identified.
Read the recording notes The agent receives a request to check the inbox for likely phishing or scam senders. It runs an mxr search for suspicious wording and reads the returned matches. This is a recorded excerpt, not a current capability or model-quality test. Open video: Have an agent search the synthetic inbox Your browser cannot play this video. Open the recording .
Let an agent compose a search pipeline 23s The capture shows an agent combining an mxr JSON search with jq to inspect message fields.
Excerpt from a recorded synthetic-inbox agent session. GPT-5.5 is visible; date and mxr version are not identified.
Read the recording notes The agent searches inbox mail and asks mxr for JSON output. A jq expression selects message details for the suspicious-sender review. The capture shows shell composition; it does not measure time saved or model accuracy. Open video: Let an agent compose a search pipeline Your browser cannot play this video. Open the recording .
Review a read-only agent answer 20s The agent reads synthetic messages, inspects headers, and gives a suspicious-sender summary without changing mail.
Excerpt from a recorded synthetic-inbox agent session. GPT-5.5 is visible; date and mxr version are not identified.
Read the recording notes The agent searches matching inbox messages and fetches selected headers. It summarizes the possible scam messages and advises verifying shipping notices independently. The closing text says it checked read-only and changed nothing; this is a recorded answer, not a live model evaluation. Open video: Review a read-only agent answer Your browser cannot play this video. Open the recording .
Search a 50,000-message synthetic mailbox 16s A CLI search for contract-renewal mail displays 50 results from 97 matches in the synthetic dataset.
Recorded CLI capture against a synthetic 50,000-message dataset. The clip shows one result page, not a search benchmark.
Read the recording notes The command searches for contract-renewal messages. The terminal reports 50 results on this page and 97 total matches. No timing comparison or performance measurement is shown. Open video: Search a 50,000-message synthetic mailbox Your browser cannot play this video. Open the recording .
Combine mail search with jq and subscription ranking 21s The shell capture pipes search JSON through jq, then runs the CLI subscription-ranking command.
Recorded CLI capture against synthetic mail. The displayed open-rate label is a local READ-state proxy, not tracked opens.
Read the recording notes The command searches newsletter and digest messages and pipes JSON fields through jq. A second command lists subscription candidates in rank order. The displayed rate is based on current READ flags and applies only to candidates returned for ranking. Open video: Combine mail search with jq and subscription ranking Your browser cannot play this video. Open the recording .
Rank returned subscription candidates by read state 17s The CLI ranks a limited set of subscription candidates using the share of messages currently marked READ.
Recorded CLI capture against synthetic mail. Its on-screen “open-rate” wording is a legacy label for current READ state, not distinct message opens.
Read the recording notes A separate search displays newsletter and digest matches from the synthetic mailbox. The subscription command ranks its returned candidates. No explicit account selector or candidate limit is shown. The numerator is the current local READ flag; mxr does not track whether a person opened a message. Open video: Rank returned subscription candidates by read state Your browser cannot play this video. Open the recording .
Search local mail in an older TUI layout 13s This historical TUI capture searches a synthetic 50,000-message mailbox and opens a matching thread.
Recorded synthetic-mail TUI capture. The panels and navigation are from an older interface and do not show the current default navigation.
Read the recording notes The search view loads results for a query across the local mailbox. The recording shows a matching result selected beside its message content. The screen is a historical layout, not a guide to the current default navigation. Open video: Search local mail in an older TUI layout Your browser cannot play this video. Open the recording .
Inspect a rule in an older TUI workspace 14s A historical TUI capture opens the Rules workspace and shows the overview for a demo newsletter rule.
Recorded synthetic-mail TUI capture. Its multi-panel navigation is historical and differs from the current default navigation.
Read the recording notes The capture opens the older Rules workspace and selects a demo rule. The overview displays a sample condition and says no action was recorded. It does not demonstrate applying the rule or current TUI navigation. Open video: Inspect a rule in an older TUI workspace Your browser cannot play this video. Open the recording .
See separate TUI, CLI, and agent recordings 35s An edited montage cuts between a TUI mailbox, a CLI search, and a recorded agent session.
Composite of separate TUI, CLI, and agent captures using synthetic mail. The edits do not show a continuous handoff or prove shared-state synchronization; the web client is not shown.
Read the recording notes The first segment shows the TUI mailbox and a selected message. A separate segment shows a shell search and its results. The final segment shows an agent prompt and search. Treat these as distinct recordings, not one continuous session. Open video: See separate TUI, CLI, and agent recordings Your browser cannot play this video. Open the recording .
A montage of mxr client screens 32s This composite cuts between TUI, CLI, and agent screens; it does not show source code or a live client handoff.
Composite of separate synthetic-mail TUI, CLI, and agent recordings. The web client is not shown, and the recording does not demonstrate the Rust implementation.
Read the recording notes The montage shows a TUI mailbox and message view. It cuts to a separate CLI search and a recorded agent search. The sequence is illustrative editing, not a continuous session or implementation walkthrough. Open video: A montage of mxr client screens Your browser cannot play this video. Open the recording .
Preview an archive selection in the CLI 12s The CLI dry-runs an archive search and lists the two selected threads and two messages before any change.
Recorded CLI capture against synthetic mail. Despite the old asset name, no agent appears in this clip.
Read the recording notes The command selects unread messages from a synthetic notes-list address. The --dry-run output says it would archive two threads containing two messages. The preview does not apply the archive action. Open video: Preview an archive selection in the CLI Your browser cannot play this video. Open the recording .
List two synthetic accounts, then search one address 14s The CLI lists two synthetic accounts and then searches messages from one demo address.
Recorded CLI capture using two synthetic accounts with fake sync and send settings.
Read the recording notes The accounts command lists Demo Personal and Demo Work. A separate search uses the notes address and displays matching rows. The search does not specify an account, so this clip does not establish combined cross-account scope or real-provider behavior. Open video: List two synthetic accounts, then search one address Your browser cannot play this video. Open the recording .
Find messages with attachments 12s A CLI search filters synthetic mail for one sender and messages with attachments, returning two rows.
Recorded CLI capture against synthetic mail. It demonstrates search results only, not downloading or filing an attachment.
Read the recording notes The query combines a sender filter with has:attachment. The terminal lists two matching messages with subjects and dates. No attachment is downloaded, opened, or added to a manifest in this recording. Open video: Find messages with attachments