Turn information noise into auditable, structured insight.

No. 100+ built-in source presets work out of the box — pick a template, add a keyword, done. If you want custom collectors or multi-step LLM orchestration, the visual editor covers that too (still zero code).

It depends on source count, article length, and the models you pick. A Chinese daily briefing over 50 sources — gpt-4o-mini for summaries plus claude-3.5-sonnet for synthesis — costs roughly 80–150 credits. Billing is per-token, with a balance pre-check that prevents overdrafts.

You do, entirely. Multi-tenancy enforces userId filtering on every query, API keys are stored AES-256-GCM encrypted, and organization visibility is fine-grained. You can export raw JSON at any time.

Yes. We provide proxied access to OpenAI and Anthropic, and first-class support for domestic models such as DeepSeek, Qwen, and Moonshot. Google Search is blocked in China, so we fall back through Jina / Tavily / Metaso.

No. LLM billing runs in atomic transactions with row locks — a failed charge rolls back or compensates automatically. A failed collection only updates its timestamp and never touches credits. Every charge traces back to a concrete call log.

Yes. Reports are delivered on schedule to Lark groups (card messages), Slack channels (Blocks format), and inboxes (HTML email with custom SMTP). The reverse direction works too: IMAP email ingestion, token-authenticated inbound webhooks, and mentioning the bot in a Lark group can trigger report generation directly.