Research messenger and AI consultation

Unread badges and receipts

1. Conversation, Research chat and AI consultation badges count currently accessible unread entries. Notification-inbox read status is separate.
2. Messages actually displayed in the focused conversation are marked read automatically. Viewing the directory or leaving older messages unloaded does not acknowledge them.
3. Open Read or the people icon beside a heart to see readers or likes. New or revised messages may need acknowledgement again.
4. Explicit Mark conversation read acknowledges its displayed boundary. Later arrivals remain unread.

Open conversations and optional metadata

Cite a page like a wiki

1. In the message box, type [[ and part of a title, such as [[Pasa.
2. Choose a grant, project, material, report, event or researcher using arrow keys and Enter, or click.
3. The input receives [[title|internal path]]. Add multiple citations and send to display clickable titles.
4. Research chat suggests only targets accessible to every current participant. AI consultation uses your own access. Search is by title.
5. Add tags with comma or Enter and remove with ×. Metadata for a new conversation is optional and separate from the currently selected chat.

AI errors and researcher activity reviews

1. For a failed consultation, inspect the explanation in Full history. Resolve permission, allowance or readiness issues before Retry. Retries also count as attempts.
2. Researcher reviews are scheduled according to the configured evidence window and refresh interval. Repeated profile reloads do not run them.
3. Profile summaries and detailed reviews have different audiences. The subject and platform administrators can read details; evidence links require current access.
4. See AI assessment for role-specific criteria, evidence periods and unassessed items. Reviews support mentoring and do not prove experimental success or objective personnel ratings.

Start: create channels and groups

1. Open the floating conversation button → Research chat → +. Optionally name it and select people, grants or projects.
2. Selecting a grant/project can include its current participants. Review the actual audience before creating. Topics and tags alone do not invite people.
3. Choose Start. A researcher's profile also offers a direct-conversation button.
4. Private groups are readable only by current participants. Nonparticipant platform administrators do not have blanket group access.
5. Close only folds the panel. Leave conversation ends participation and subsequent access and notifications.

Manage, search and download history

1. Open Profile → Settings → Conversation management or Full history in the panel. Search names/content and filter by type, dates and context.
2. Open chat opens the side panel; Conversation settings manages the name, participants and automatic summary. More separates download, clear, leave and trash actions.
3. TXT/JSON exports include up to 2,500 entries. Split larger periods and download attachments through their links.
4. Only you manage your private AI consultation. Group administration still requires current participation and applicable authority; nonparticipants cannot bypass privacy.
5. Clearing, trashing and leaving have different effects. Review the confirmation. They do not jointly erase other people's preserved messages, existing reports or external copies.

AI titles, commands and read-only sharing

1. Authorized AI users click + in AI consultation and optionally select context metadata. The title is optional: leaving it blank uses the beginning of your first question. Rename it in Name and settings.
2. Type / to see commands. Typing /report w narrows the list. Select with arrow keys and Enter or the mouse. Autocomplete does not send anything; review before sending.
3. Use /report daily, /report weekly, /report monthly, or two dates. Replace YYYY-MM-DD placeholders with real dates. Report permissions and usage limits still apply. In a new private consultation, use the Reports page to choose generation conditions. Existing grant/project consultation commands remain available.
4. AI conversations are private by default. Enable Share read-only with channel participants in settings to share the history link with that channel. Readers cannot ask, edit or delete, and answers are hidden when they lack evidence access. Disable the option to revoke sharing.
5. Use the small share icon on the history page to copy its link or share to another channel/app. A link never grants access. Current research channel messages are searchable AI evidence; private AI answers are not independent proof of research progress.

1. Chat while browsing

1. Click the floating conversation icon at the bottom right to open the right panel. Toggle it again or press × to close. On mobile it uses the full screen.
2. Open a conversation from the list or press + to start one. Grant/project rooms are not automatically created. Navigation restores the conversation selection and typed draft in the same browser tab. Unsent file selections must be reselected after navigation.
3. Type and send a message. Type @ followed by the start of a username to select a participant. Mentions and replies create notifications for the recipients.
4. The open panel refreshes about every five seconds without forcing you to the bottom while reading older messages. Choose View history for full-channel search, editing and exports.
5. Grant participation does not automatically grant access to every child-project channel. Obtain approval for the relevant project.

2. Replies, search, citations and sharing

1. Choose Reply / details to open a stable message URL. Replies notify the original author inside the platform, and mentions notify selected participants. Email requires administrator Gmail setup and recipient opt-in. SMS is not provided.
2. Enter words in channel search. Expand advanced search for author, tags and date boundaries.
3. Copy a citation link and paste it into another conversation. On a report, material or event, choose the small share icon → Share to channel → target channel. Review the prefilled link and message before sending.
4. Recipients must independently have access to the linked page. Sharing does not fetch or preview unauthorized source content.

3. Edits, deletion and confirmed actions

1. Open Edit / delete on your own message, or on a message in a scope you manage. Save the edit or delete the message. Refresh if someone changed it concurrently.
2. Deletion leaves a tombstone; earlier versions are retained. Editing clears the previous decision confirmation. Authors and managers can inspect revisions.
3. Managers can mark a confirmed decision, confirmed action or unresolved question. Discussion does not establish experimental success.
4. A confirmed project action offers links to connected GitHub boards. Review the proposed issue and explicitly confirm the remote change. A stable request identifier prevents repeated creation from the same message. If a network outcome is uncertain, check the remote issue first.

4. Grant AI conversation access

1. A platform administrator opens Profile → Lab management → AI access and daily limits. Being a grant or project manager does not grant authority to enable AI access for others.
2. Select a user, enable AI conversation access, choose allowed models and maximum reasoning effort, then save. Choose from the available models and limits shown in the form. Use Change on an existing permission card to edit it.
3. To revoke, clear Enable and save. Queued or running requests are invalidated. Owners can still read their completed private answers while they retain conversation access. New requests and shared answers require current permissions.
4. The daily AI allowance is shared by reports, translations and consultation. Enabling chat does not expand data access or report management rights. Actual model availability depends on the server Codex account.

5. Consult AI and request reports

1. Authorized users choose Conversations → AI consultation → +, optionally add a title and context, then save. Private consultations are excluded from shared lab Drive archives.
2. Ask for confirmed progress versus unresolved issues, a researcher’s evidence-backed activity, paper experiments or future research directions. Expand Model and effort to change settings for each question.
3. Wait for processing and use the new-response notice to refresh. Expand citations and compare the evidence. Retrieval searches readable materials, messages, events, reports and your own consultations for relevant excerpts; it does not claim to read all originals. Collect external materials before asking about them.
4. If you manage the scope, expand Generate research report in this conversation and choose daily, weekly, monthly or a custom range. You can also type /report daily, /report weekly, /report monthly or /report 2026-09-01 2026-09-13. The report link appears in the conversation and in Reports. Equivalent requests reuse existing reports.
5. Ordinary natural-language consultation does not change settings, send messages or execute experiments. Use the explicit report command or control for creation. Your completed answers remain in history; shared answers, source links and new requests follow current access. Do not treat an old answer as current evidence; ask again when needed.

6. Files, exports and clearing history

1. Research chat and AI consultation support file selection, pasted images and dropped files: up to three files, 1.5 MB combined. Review the list before sending; see Compose and attachments.
2. Channel files require current participation; private consultation files follow that consultation's audience. Links grant no access.
3. Export history by dates, author and keywords as TXT/JSON. Download original files through their attachment links.
4. Clear/delete follows the confirmation scope. Deleted messages are excluded from new AI retrieval; existing reports and historical copies are separate.
5. Private consultations, direct messages and participant-only summaries are not all copied to the shared lab Drive. Operational DB recovery backups are distinct from app sharing.
6. Direct messaging is available; voice/video calling is not. Use a separate Google Meet event for meetings.

Mentions, replies and daily limits

1. In research chat, type @username, choose a participant and send. Participants who can currently read the channel receive notifications. An explicit self-mention also creates a notification; ordinary messages you send do not notify yourself.
2. Click @username to open the profile, or ↗ next to a reply to open its original message. Code blocks and email addresses do not count as mentions.
3. In private AI consultation, @username is a reference only. It does not share the conversation or notify that person. Use research chat for shared communication.
4. Platform administrators: Profile → Lab management → AI access and daily limits → Change on a user card. Set AI conversation access, daily attempts, allowed models and maximum effort, then Save. Selecting another user reloads their saved settings.
5. Reports, AI consultation, AI translation and normal material analysis share the daily attempt allowance, including failures. On-device translation uses none of this allowance. Blank uses the role default; 0 disables usage. The counter resets at midnight in Seoul. Token-based limits are not provided. Enabling AI does not grant additional grant or project access.

Free message translation, originals and reuse across devices

1. Use A↔가 below a message to translate between Korean and English. Switch to Original or Both to compare. Links, code, mentions and attachments remain intact.
2. Choose interface language, Korean, English or Always original in translation preferences. Your preference and private copies sync across devices. Reader translations remain private; only author-approved shared translations appear for other current participants.
3. Expand Translation preview below the composer, translate for free, review and opt to share with the original. Optional previews while typing never block sending the original. Recheck after editing.
4. Desktop Chrome uses the Translator API without consuming AI allowance. First use may require clicking the icon to download a language pack. Unsupported browsers including mobile display saved copies. Create a new copy on desktop or use Chrome menu → Translate yourself. The website cannot trigger or read Chrome menu translation.

Read recent conversation and earlier history

1. A conversation opens on recent history. Expand earlier conversation or Load older messages to read more.
2. While browsing older messages, Show latest returns to new arrivals. Full-history search also searches folded content.
3. Automatic receipts cover displayed messages; Mark conversation read has a wider explicit boundary. Unseen older messages are not automatically acknowledged.
4. A message link opens its surrounding history after an access check. If unavailable, check current channel participation.

Find unread messages in one place

1. Use the conversation badge or unread list to choose a conversation. Research messages and new answers in your AI consultations are distinguished.
2. Search and paginate the full unread directory for more; ordinary conversation lists also prioritize unread activity.
3. Displayed messages in a focused panel receive automatic receipts. Closing, listing or receiving push alone does not mark unseen messages read.
4. Network conditions may delay refresh. Notification read state, conversation read state and pending approval counts are separate indicators.

Direct messages, lab channels and prior consultation

1. Start a direct message from a profile or choose a person in the conversation picker. Continue an existing conversation when one opens.
2. Use an existing lab channel or invite the intended participants to a new conversation. Check the channel and participants before sending something private.
3. AI consultation can retrieve accessible collected documents, GitHub Projects work, conversations and your own completed consultations. A connection does not mean unlimited reading of everything it contains.
4. Name the project, period and document and cite relevant sources. Open the answer's references to verify coverage and omissions.
5. Other people's private consultations and support inquiries are not research conversation evidence. Completed visual analysis of a PDF/image can also supply supporting notes.
6. Click the copy icon below an AI response. After Copied appears, paste the answer into your document.
7. If automatic copying is blocked, copy the selected response manually. Answer copying excludes the question and separate source list.
8. The history page copy icon copies both the question and answer.

Manage participants and periodic AI summaries

1. Check current participants in Conversation settings. The creator can remove other participants, ending their access and alerts. Leave ends your own participation.
2. Review automatically included scope participants when creating the conversation. Tags do not automatically rejoin someone who left.
3. Enabled summaries become eligible roughly every 30 minutes when new messages exist. A current participant's AI permission, lab settings and allowance are required; completion is not immediate.
4. Open AI conversation analysis to inspect status and the accumulated summary. Pausing retains valid results. Edited or deleted originals may require reanalysis.
5. Summaries are participant-only AI interpretations, not confirmed decisions or research results. The summary task does not automatically read the full originals behind attachments or citation links.

Compact message translation and deletion

Zoom, pan and download in the image viewer

1. Select an image thumbnail to open the viewer. Use zoom buttons or the wheel on desktop and pinch gestures on mobile.
2. Drag an enlarged image to pan. For multiple images in one message, use previous/next. Close or press Escape on desktop to return to the conversation.
3. Choose Download separately in the viewer. It downloads the stored image through its authorized path and does not grant access to private attachments.
4. On small screens use controls within the safe area. Zooming cannot recover detail absent from an older low-resolution stored image.