Plugins
Meetings
Keep notes, transcripts and reviewed outcomes together.
Keep meeting notes, imported transcripts and reviewed outcomes together. Meetings does not record calls or infer decisions. Your own notes are separate from imported sources and survive source refreshes.
Capture and read
Create a meeting under /account/meetings, then add notes or import an UTF-8
TXT, Markdown, WebVTT, SRT or Meetings JSON transcript. File previews show an
excerpt; the import retains the full original, including its BOM and line
endings. Invalid encoding or malformed captions are refused explicitly.
A Markdown file exported from Jamie, a Google Meet notes document or a Wispr Flow export is read by a converter: every line keeps its speaker and its time, and the file's summary is imported as a second source of the kind summary. The import preselects the converter by the file's shape and you confirm or change it. A converter refuses a file it does not recognise instead of reading it as plain paragraphs. Each source shows its kind: transcript, summary or notes.
A source starts private. Once its import is complete, its author may share it with people who can read the meeting. Sharing a meeting alone does not share its private sources. Revoking a source's sharing also removes access to its transcript parts. Partial imports are labelled and never presented as complete. Older source generations remain available as immutable evidence.
Outcomes
Decisions, questions and action proposals start as proposals. A person reviews and confirms them in the browser. Transcript text and provider summaries are untrusted evidence; neither can approve an action or change permissions. Confirming an outcome releases its reviewed text to the meeting's readers. It does not release its source transcript. Tasks owns actual work; an action proposal is not a second task lifecycle.
MCP and portable input
Load plugin_context("meinkontext.meetings") before using the workflow.
meetings_query searches authorized records; calendar_event_id narrows it to
the meetings linked to one Calendar event. meetings_transcript reads a
bounded page and returns the source revision/version for citations.
meetings_import imports a source into the meeting named by meetingId, or
finds or creates the meeting itself when meeting carries its fields and an
intakeKey; a repeated identical call creates nothing. Its converter names
a file converter (jamie.markdown, google-meet.notes-document,
wispr.flow-export), its artifact the source's kind. A source records the
converter and version that derived its segments, and completion derives them
again with the same version. Use plugin_command with action
meinkontext.meetings.command to create meetings or propose outcomes, passing
expected revisions and an idempotency key.
Source identity is provider + account binding + external ID + artifact within
the workspace. Hashes detect changes; they are not identities. Repeating the
same import returns its existing source. Replacing changed content requires
expectedSourceRevision; meeting notes and confirmed outcomes are not replaced.
Meetings JSON is { "schemaVersion": 1, "segments": [...] }. Each segment has
id, text, speaker (string or null), and startMs/endMs (nonnegative
milliseconds or null). Segment IDs must be unique and end cannot precede start.
The optional providerPayloads array retains selected transcript content from
a provider response, excluding transport metadata, credentials and signed URLs.
Generated schema files in contracts/ are authoritative.
Provider imports also use schemaVersion: 2: each entry in pages contains
segments in the same segment format and a providerPayload string. This lets
readers process provider content one page at a time. Page order is preserved;
segment IDs remain unique across pages. Google timestamps are relative to the
earliest entry for individual imports and the known conference start for
scheduled imports.
Optional connections
Connections and credentials belong to Integrations. Connecting an account does not import history, enable polling or share transcripts. Google Meet uses its read-only conference/transcript scope. Jamie uses its official remote MCP endpoint; approve meeting read tools individually. Wispr supplies Notetaker content, not dictation history. Provider transcript access may be narrower than summary access. Disconnecting stops reads but does not erase accepted local copies or revoke authorization at the provider.
Provider live acceptance requires synthetic test accounts. See the implementation checklist for current readiness; the complete provider import and scheduling workflow is still under development in the single implementation PR.
File upload, refresh and history
The browser uploads the original in parts. Pause stops after the current part; select the same file to resume. A partial source remains labelled partial and private until finalization succeeds. The current finalizer still verifies all parts together; large-source bounded finalization is an open implementation item.
Choose an existing file source explicitly when replacing changed content. For provider imports, selecting an existing source also confirms the account mapping if the connection has changed. The provider and external source identifier must still match. New connections start with separate account identities; names and email addresses never merge them automatically.
Download checks the complete original's SHA-256 before offering a file. History uses the host's additional history permissions; current read access alone does not grant access to previous revisions. Source speaker corrections are separate labels and optional Contact references, leaving original labels unchanged.
Reviewed remote field mappings
For Jamie and Wispr, select an approved read operation and map its actual returned
shape. The following is a synthetic example, not a claim about either live
provider's schema. Given a response with transcript, next and done fields:
{
"content": ["transcript"],
"kind": "text",
"meetingInput": "meeting_id",
"cursorInput": "cursor",
"nextCursor": ["next"],
"completion": { "kind": "flag", "path": ["done"], "doneValue": true }
}For arrays, use kind: "segments", idField, textField, and optional
speakerField, startField, endField, timeUnit (seconds or milliseconds).
Only selected original fields are retained; normalized timing is a separate
derivative. Missing completion flags and contradictory cursors fail explicitly.
Use completion: {"kind":"cursor-empty"} only when the provider documents that
an empty cursor proves completion. Schema changes require a new operation review.
An automatic rule's advanced bindings contain list and transcript. Each
binding has connectionId, operation and the approved schemaDigest shown by
Integrations. transcript.mapping is the mapping above. list.listing describes
the actual list response, for example:
{
"items": ["meetings"],
"idField": "id",
"titleField": "title",
"startedAtField": "startedAt",
"sinceInput": "since",
"cursorInput": "cursor",
"nextCursor": ["next"]
}This example requires ISO timestamps and reliable date-filtered enumeration. Do not enable polling for a search-only operation with incomplete enumeration. Bindings in the form must use the selected connection. Google bindings are assembled from approved operations. Drive requires an explicit document query.
Automatic checks and follow-up
Rules are opt-in and stay in the selected workspace. After the initial selected history is processed, checks revisit a seven-day overlap; older edits need a manual reimport or an explicitly changed backfill selection. Editing a rule resets its scan when selection or bindings change. The overlap stays fixed while paging. Failures retain the cursor and show a safe status; no provider response is copied into error messages. Rule creation alone does not enable the server's API import consumer; its activation is a separate operator configuration.
The follow-up briefing lists this meeting's confirmed decisions, unresolved questions and action references deterministically. Linked Tasks are resolved under current rights. Missing Tasks access produces an unavailable reference, not a cached task title. Search covers accessible transcript parts and outcomes, with record revision and import generation. It may find historical generations; it does not present them as the current source.
Meeting export and archive
“Export meeting” downloads JSON with the notes, outcomes, source data, originals and normalized transcript parts currently visible to you. It includes versioned citation references, never credentials or execution payloads. Original pages are checked against their SHA-256; a failed read or changed source aborts the download. Incomplete imports remain labelled and do not claim a complete original. Older originals are available individually through source history. Archiving hides the meeting and its sources from readers; restoring preserves originals and outcomes.
Permanently remove a local source
The source's importer may first archive it. “Archived sources” offers restore or permanent deletion after a separate confirmation. Deletion removes every local original and transcript version. Confirmed decisions and independent Tasks remain; deleted evidence no longer resolves. Another meeting editor does not gain authority to erase your sources.
The same source controls offer an optional deletion deadline in UTC. Confirm it explicitly; without a deadline the source remains indefinitely. Clear the date to cancel the policy. Archiving does not cancel a deadline; restore an archived source first to change it. The worker discovers due sources every minute and rechecks the live policy, revision, ownership and rights under the host lock. Deletion may be delayed by maintenance, disabled plugins or lost permissions. It uses no provider credentials and never deletes anything at the provider. Imports and refreshes cannot opt you into a deadline; refreshes preserve a policy you already chose. Incomplete staging imports are cleaned up automatically after 24 hours.
Provider imports first show a private excerpt (up to 4,000 characters), total segment count and size. Confirm that you may keep a local copy, including for material shared by someone else. Confirmation reads the provider again under current access and compares the full content digest; a change requires a fresh preview. The excerpt limit does not truncate the imported transcript. This preview stores no local Meeting source. Recurring rules remain a separate opt-in.
Automatic remote imports additionally require owned in both mappings: a
nonempty path to a provider-documented boolean meaning owned by the connected
user. The list path is relative to each item; the transcript path is relative
to each response page. False listing values are skipped; missing or nonboolean
values stop processing. Ownership is checked again on every transcript page.
These paths must be reviewed against the actual operation; ownedByMe is only
an illustrative field name, not a promised Jamie or Wispr API field. When the
provider exposes no reliable ownership signal, use individual import with copy
review. Do not map mere read permission or shared access as ownership.
Drive rules constrain listing to 'me' in owners and recheck ownedByMe before
exporting each file. Google Meet's conference listing is organizer-filtered
according to its official contract.
Provider-backed imports retain the approved read binding for hourly access checks on the opt-in API read-job service. These checks mark a confirmed missing, trashed or forbidden provider source unavailable; rate limits, schema changes and connection problems show a failed-check status instead. Accepted local originals remain readable. A successful later check restores availability. Drive checks metadata only; remote providers check the first transcript page. Disabling the connection stops reads. Archived sources are not polled. Check status is not a promise of instant provider ACL mirroring.
Incomplete imports have a 24-hour resume window. After that, the keyless worker removes the unfinished generation in batches. An interrupted refresh restores the previous complete source and preserves its original files and citations. A new source that never completed is removed. This cleanup does not remove independent notes, decisions or Tasks. Completed sources keep the retention policy you chose.
Meetings open in reading view. Use Notes, Outcomes and Transcripts to change the working area; Edit opens the note editor. Unsaved text stays available when changing tabs. Provider setup uses approved-operation selectors and labeled mapping fields; field paths use JSON Pointer notation, with an empty path selecting the full response.
For an approved operation verified to return the entire transcript in one response,
select the single-response option. Its mapping explicitly uses cursorInput: null
and nextCursor: null; no made-up cursor field is required. Do not use this for
search snippets or truncated responses. If a completion flag is mapped and reports
incomplete, import still fails rather than accepting a partial transcript.
In the browser, select provider sources before importing: Google Meet lists conference occurrences and then completed transcript artifacts; Google Drive searches document titles and modification dates. Jamie/Wispr use the approved listing operation and its reviewed field mapping. Selection does not write any meeting data. Review the preview and explicitly confirm the private copy next. When listing is unavailable, direct source IDs remain under the details. Automatic intake still requires owned sources; a manual copy of a shared source requires your explicit confirmation.
Scheduled Meet, Jamie and Wispr imports save content pages individually and resume from saved progress after an interruption. Pages are reread and compared before publication. Observed changes leave the import incomplete; without a provider snapshot contract this does not promise an atomic upstream snapshot. The UI shows fetch, verification and final-check phases. Use “Refresh status” to reload progress.
“Discard import…” lets you confirm early cleanup of an incomplete source. Cleanup runs in the background and removes only the incomplete generation, preserving previously completed originals, notes and outcomes. Another import of the same source waits for cleanup. Unchanged scheduled refreshes are likewise cleaned up without publishing another accepted content generation.