Cron runs (newsletter generation jobs triggered by the schedule) and newsletter sends (each batch sent via Mailgun). Click any row to expand details — failed sends, per-slice timing, recipient lists.

Used to personalize copy on your public share pages — e.g. the share landing says “Who would you like to meet from Ion's network?” when your first name is set.

Anyone who signs in via a domain-restricted share link is added here. Admins have full access to this app; regular users can only see their own network + any shared networks. Admins listed via the ADMIN_EMAILS environment variable always have admin role.

How the Network tab and all share pages handle search queries. Deep uses Diameter's agentic search — it discovers relevant companies in your network, runs targeted searches, and verifies borderline matches. Significantly higher quality on industry / multi-hop / ambiguous queries, but takes 30-60 seconds per search and costs ~$0.05-0.15 in API credits. Fast uses the legacy filter+embedding pipeline — 2-3 seconds, ~$0.005 per search, but misses industry-bridging matches. End users on share pages never see a toggle; this setting controls behavior for everyone.

Which API to call when researching a contact via the Network tab's Research buttons. Waterfall modes try the primary provider first and fall back to the secondary one if the primary returns no useful fields.

How much of your LinkedIn network has been researched via Exa (structured fields populated) and embedded for semantic search. Click the action buttons in the Network tab to backfill.

Scheduling is handled by a Render Cron Job — edit the cron expression in your Render dashboard. The cron hits POST /api/cron/trigger with the shared CRON_SECRET. When triggered, the digest is always generated; the toggle below controls whether it also auto-sends to subscribers.

When on, subscribers can reply to the newsletter email and Dia will respond using the newsletter contents + web search. Replies are routed via the same dia@ address; dispatch happens by the In-Reply-To header so newsletter ingestion is unaffected. Guardrails: only active subscribers, 5 replies/24h each, 10 turns per thread.

Sentences each story summary should aim for, set separately for Top News (significance ≥7) and the rest. Setting both to 1 reproduces the original behavior exactly. A common pattern is 2-3 for Top News + 1 for the rest: richer headlines, scannable rest of the digest.

When on, anyone can visit /subscribe to sign up. They'll receive a confirmation email and only start getting newsletters after clicking the link inside. Turn off to close public sign-ups (admin-added subscribers in the list below are unaffected).

Categories Claude can assign to each tracker post, in the order they appear in the digest email. "Other" is appended automatically as a catch-all and isn't shown here. Removing a category silently rebuckets older posts to Other.

When on, each newsletter generation pulls top HN stories (≥100 points) and runs them through a Diameter VC-relevance filter (rubric: kernel hacks and dev trivia get 1-2, founder essays and industry news 7-10; we keep ≥6). Top-scoring items appear in a dedicated From Hacker News section. Adds ~5s and roughly $0.005-0.015 per generation.

The newsletter reads X posts from a cache populated by a separate fetcher cron (hit /api/cron/x-posts-fetch every ~6h). That way 12h and 24h newsletters share the same source of truth, and Grok latency at generation time goes to zero. Each fetch costs ~$0.05-0.20 per enabled angle. Requires XAI_API_KEY on the server.

Blogs ingested alongside the news feed each generation. RSS sources are fetched directly; llm_scrape sources have their index page parsed by the model (slower, costs ~$0.03/source/run, works on sites without an RSS feed).

Emails on this list receive the newsletter when you click Send or when the daily schedule fires (with auto-send on).