scene 02, ai tooling
Four agents you can poke. Each runs the same way I'd wire one into a client's system: human-in-the-loop, no auto-publish, provider-agnostic via AI Gateway.
Drafts blog posts on-demand via Anthropic. Streams tokens straight into the panel. Drafts land in a review queue, never auto-published.
Sweeps the audit log every night for spikes, DELETE bursts, lockout cascades, stuck states. Emails a digest if anything trips.
Takes inbound submissions and classifies them as legit / suspicious / spam, with the reasoning surfaced for human override.
Reads a draft against your style guide, flags voice mismatches, suggests edits inline. Designed as a second opinion, not a publish gate.
ECHO drafts content in Qamar's voice via streaming Anthropic tokens. PULSE sweeps the demo's data for anomalies, same SQL rules that fire nightly in production. Each run logs to the activity feed below.
click run echo to draft
Runs the same anomaly sweep that fires nightly in production. No AI - pure SQL over the audit log, leave requests, inquiries, and tasks.
click run scan to sweep the last 24h.
paste or pick a sample, then classify.
paste or pick a draft, then review.
⚠ ECHO + RIVER + ATLAS come fully live the moment AI_GATEWAY_API_KEY is set · PULSE is SQL-only and works now