3. Getting knowledge in#

Upload (Workspace tab): PDF, Word, Excel, CSV, text/Markdown, HTML — and scanned images (JPEG/PNG/WebP/GIF, up to 5 MB, transcribed by the vision model; requires image processing + an Anthropic key, §8). During upload you can set the batch's groups and tags in one step. Batches up to 60 MB, 25 MB per file.

Connectors (Connectors tab): define, enable, and pre-configure the input and output modules available to the workspace. Input modules include S3-compatible buckets, Google Cloud Storage, Git repositories (GitHub etc.), Google Drive, Notion, SharePoint / OneDrive, Confluence spaces, Zendesk Help Center, Linear, and Slack as a companion app. Disabling a connector type blocks new sources of it; existing ones keep syncing. Connector cards show enabled/disabled state, active source count, configured/unconfigured state, setup requirements, and a direct link into Workspace with that source type preselected.

Sources (Workspace tab): select the enabled connector, create or connect the source, set default groups/tags, and run the ingest flow for that source. Re-crawling only processes changed content. The Sources list is split into a compact scan row and an Edit panel per connected source. The row keeps the important operational actions visible: source name, status, document count, recent activity, Crawl now, Edit, and admin-only Delete. Edit opens the connector-specific configuration fields plus sync cadence, auto-curate, and content shelf-life controls. Secret fields are write-only: leave them blank to keep the stored encrypted credential, or enter a new value to rotate it. Sources can be scheduled to sync automatically (hourly / daily / weekly) or left manual. Workspace also opens with an ingest cockpit: active source count, running/failed jobs, pending curation, re-verification attention, recent source changes, and recent job failures.

Output connectors (Connectors tab): send this workspace's LLM traces to your own Meridian account. Paste the key pair from your Meridian org and click Activate — a test trace validates the credentials before anything is stored. Traces are metadata-only (model, feature, tokens, latency, status) unless you opt in to include prompt/completion content. Pause stops tracing and keeps your keys; Deactivate wipes them.

Agent memories (§9): your team's agents can deposit distilled facts.