Data and indexing

Collections

/dashboard/collections lists the organization's collections with health at a glance: document count and share, capability badges for searchable, faceted and semantic (vector-bearing) collections, and a "needs attention" signal derived from the schema. Stat tiles count collections, documents and indexed fields; the fourth tile reads Need attention when any collection has an empty index, no searchable field or no filters, and Semantic-ready (with a hint) otherwise.

• Search and sort by name or field, newest / name / most documents / needs attention. Press / to focus the search box and Esc to clear it. The sort choice and a table-or-cards layout are both remembered per browser.
• Row actions open the collection, browse its documents, preview search, edit the schema, copy the name or schema JSON, and delete.
• Creating, editing and deleting are owner/admin actions. A member sees the buttons disabled with the reason, rather than hidden.
• Deleting a collection drops its documents and disables search rules, synonyms and curation sets that name it. The confirmation says so; the action cannot be undone.
• The empty state offers New collection and Guided setup (the quick start).

Create a collection

/dashboard/collections/new is a three-step wizard:

Collection detail

/dashboard/collections/{name} shows the schema, a document preview, a retrieval testing panel, ready-made code samples and import. Tabs are mirrored in the URL (?tab=). Stat tiles cover documents, fields (with the searchable/faceted/sortable hint), creation time and default sort, plus nested fields, vector-field count and separators. A problem list appears above the tabs when the schema would not search well, or a note that semantic and image search can run on this collection. Actions open the documents, the schema editor, the search preview, the search playground, the query graph, the delete confirmation, and copy the name or the schema JSON.

• Schema — a field explorer with a "find a field or type" box, multi-select trait filters (searchable, faceted, sortable, optional, vector, needs a look) with live counts, collapsible prefix.* groupings, and an "In the data" column from a facet probe (value count, a 50+ cap, coverage and values per document, "never populated"). Each row can copy its field name or filter the document browser by it.
• Documents — a search box over the schema's query_by fields, the "found · search_time_ms · showing the first 20" line, expandable rows showing raw JSON, and an Open in Documents handoff that carries the query.
• Code samples — four language tabs (cURL, TypeScript, Python, PHP) with copyable snippets for creating a collection, importing documents and searching.
• Import — a JSONL drop box (paste, drag and drop, or choose a file, up to 8 MB) with an on-conflict action (upsert / create / update / emplace) and client-side line validation that blocks the import on parse errors, a missing id or an undeclared field. This is not the six-tab import dialog from the document browser; it feeds one collection from pasted JSONL.

Edit a schema

/dashboard/collections/{name}/edit applies PATCH /collections/{name}. It has two halves:

• Existing fields — name, type, flags and locale, with a Drop checkbox that strikes the row through.
• New fields — the same field editor, starting empty.

Changing a field's type or flags is expressed as a drop plus a re-add under the same name, in one request. The review card shows the exact PATCH body and blocks saving on validation errors — including dropping the default sorting field without re-adding it. Dropping any field asks for confirmation first, unsaved changes are guarded, and the save fails with an inline error if the API rejects it.

Documents

/dashboard/documents is the document browser for one collection. The engine has no "list everything" endpoint, so the screen searches with q=* and layers filters on top.

• Table — a sticky select-all column, the document id, the chosen data columns, and per-row actions. Cells render image thumbnails where the field looks like an image and tooltips for long values. Sorting is offered only on schema fields that are sortable and not arrays.
• Filters — facets measured from the data (up to 8 live facets per query, 20 values each; pinned chips are remembered per collection), a raw filter_by expression box, and an insights panel that flags dead, constant and identifier-like facets, and fields with too many distinct values to use as a filter.
• Columns picker — search and fill-rate bars per field, with a Smart reset to the default selection.
• Pagination — 10 / 25 / 50 / 100 / 250 per page, with the range, the active filter count and the engine's search_time_ms in the footer.
• URL state — collection, query, facets, raw filter, sort, page and per-page round-trip through the query string, so a view can be shared or bookmarked. ?pin=field is a one-way deep link that opens the screen with that facet in the filter bar; the chosen pinned set is kept in the browser, not written back to the URL.
• Density — a compact-rows toggle, and an insights panel whose open state is remembered.

Import

The import dialog has six tabs:

Every tab writes an ingest job (POST /ingest-jobs) that the screen polls every two seconds, showing status, fetched / imported / failed counters and the first errors. Small batches of up to 100 documents with string ids also go through the asynchronous path; larger or "dirty value" payloads use the synchronous POST /collections/{name}/documents/import instead.

Import options are the write action — upsert (default), create, update, emplace — and how to treat values whose type does not match the schema: the engine default, coerce_or_reject, coerce_or_drop, drop or reject.

The header also has an Ingest via API modal with cURL, TypeScript and Python equivalents of the same import (action=upsert) and a link to Settings → API keys, for scripting the import instead.

Export

• Export all downloads the collection as JSONL, optionally filtered, and reports how many lines were written.
• Export selected downloads just the selected documents by id.
• Delete selected removes them in one request and reports how many were deleted.
• A row can be copied by id, deleted, or opened into a document drawer with per-field rendering, a View raw JSON / Copy JSON toggle, and its own delete.

Ingest jobs

Below the browser, a panel lists the 30 most recent ingest jobs for the collection — job, collection, source, status, created, fetched, imported, failed. It refreshes every three seconds while any job is queued or processing, lets a failed job be retried, and can delete a job row (the indexed documents stay). Yandex Market jobs also show their resume cursor and pass count.

A 429 from the quota renders a dedicated "quota exceeded" state with the reset time instead of the table.

Sources (connectors)

/dashboard/sources lists the connectors that keep collections filled, with problems first. Each card shows a status badge, the provider and target collection, the sync mode and cron, the last sync (or "never synced"), the last error, a masked secret, a disabled badge when the connector is off, and a quiet badge when an enabled source has not synced for over a day and reports no error. A per-card action opens the documents this source writes.

Counts double as filters (All / Error / Syncing / Paused), each disabled when its count is zero; a free-text filter appears once there are at least five sources (/ focuses it, Esc clears it). A header Refresh re-reads the list, a failed load shows a banner with Retry, and the whole list refreshes every four seconds while anything is syncing.

/dashboard/sources/new binds one provider to a collection:

• Name and Collection — the collection field links to create one first if it does not exist yet.
• Provider — bitrix, bitrix24, prestashop, woocommerce, yml, sitemap or custom.
• Sync mode — push (the CMS posts to the connector webhook) or pull (the platform fetches a URL with a token).
• Pull URL and pull token — shown in pull mode; the token is a password field and is the module option pull_token, not the tenant API key.
• Schedule — a cron expression; the hint states that */15 * * * * and 0 3 * * * are the supported schedules.
• Enabled, with Cancel, and required-field validation on submit.

On creation the signing secret is shown once, together with the connector id and the webhook URL ({API_URL}/v1/connectors/{id}/webhook) that a CMS module posts to, each with a copy button and Open source / Back to list actions. CMS modules authenticate with HMAC headers, not an API key.

/dashboard/sources/{id} edits the connector and shows its history. A status line reports the provider → collection, the last sync (or "never synced") and "updating…" while a sync runs, beside the status, id and masked-secret badges. The header offers Refresh, Open documents (the browser filtered to the connector's collection) and Delete, whose confirmation states that the sync history goes while the indexed documents stay.

• The form covers name, sync mode, schedule, the pull URL and token (a blank token keeps the stored one) and the enabled switch. The webhook URL is read-only, and a stored error is shown inline.
• Jobs — the last 30 connector jobs with status, kind, created time, stats (accepted / applied / deleted / failed / fetched / imported) and error; a row opens a drawer with the finished time, computed duration, every stats key/value, the error and copy buttons for the job id and the stats JSON.
• Journal — the last 50 documents the connector touched, with external id, document id, deleted and updated timestamps, and Load more.

The screen polls every three seconds while the connector is syncing or a job is queued or processing, and a poll never overwrites an unsaved form edit.

Deleting a connector removes its sync history but leaves the indexed documents in place.

Integrations

/dashboard/integrations is a catalogue rather than a data screen. It groups:

• CMS stores — module-based connections, see below.
• Migrate from Algolia — the importer.
• Live data sources — the providers that already have a connector, each carrying a "Module: status" or "Connector API" badge and an Add in Sources action.
• Available now — nine tiles covering a 1C-Bitrix module, the Python, PHP and Go SDKs, the MCP server, the REST API and the widgets, each with its own copyable snippet.
• Request access — OAuth connectors that are not live yet (Notion, Google Drive, Dropbox, OneDrive, Box, Confluence, Intercom, SharePoint, Zendesk, S3, Cloudflare R2). They are deliberately marked unavailable and open a pre-filled request email; the page header carries a general Request a connector link as well.
• Built-in importers — a card linking to Documents, where JSONL, JSON and CSV can be pasted or uploaded.

The catalogue is static: it does not read your organization's connector state, so treat a tile as "this integration exists", not "it is connected".

CMS stores

/dashboard/integrations/cms connects a CMS module with one installation key and publishes per-store search settings. Its header has a Refresh and links to Subscription and payments and Sync history.

Migrate from Algolia

/dashboard/migrate-algolia is a four-step, one-shot importer — Upload → Review → Apply → Done — with a Quick start link in the header.

The wizard is honest about what cannot be converted. Nested or multi-path searchableAttributes are skipped; a synonym needs at least two terms (multi-way) or an input/synonyms pair (one-way); altCorrection becomes a one-way synonym; placeholder and unknown synonym types are skipped; disabled rules are skipped; only a rule's first condition is mapped; optionalFilters are not mapped; a redirect is stored as curation metadata only; unmappable rules are skipped.

Next

• Search and relevance
• Yandex Market YML feeds
• API keys, scopes and credits