Public-facing release notes. New skills, features, fixes.
Multi-LLM router expanded — DeepSeek + Cerebras
2026-05-09v0.116
·OpenRouter integration unlocks DeepSeek V3 + 200 other models via a single US-hosted gateway. Used for cost arbitrage on bulk classification tasks (lead_classifier, prompt_enhancement, competitor_diff) — wire cost drops 90-95% on those routes.
·Cerebras integration adds Llama 3.3 70B at 1500-2500 tokens/sec — used for low-latency streaming UX (chat widget, voice agent live responses). 20-30× faster perceived response than Sonnet for short answers.
·Router auto-picks the right model for the task: cheap_bulk → DeepSeek via OpenRouter, streaming_low_latency → Cerebras Llama, default routes unchanged.
·Both providers are US-hosted — DeepSeek prompts route through OpenRouter's US infrastructure, not Chinese servers.
Cleaner cost language + LinkedIn deferred
2026-05-07v0.84
·Every paid-action confirm dialog now shows a clean credit cost ($X.XX) — replaced the old verbose API-cost copy with plain pricing.
·LinkedIn hidden from Ad Audiences UI until the LinkedIn Marketing partnership lands. Lib + DB path preserved for when it ships.
Drill-down pages + final audit pass
2026-05-07v0.83
·New drill-downs: /aeo/[id] (citation matrix per model × query), /aeo/proposals/[id] (apply rewrite), /competitors/[id] (URL change history), /competitors/counter/[id] (counter-campaign review).
·SSRF guard centralised in src/lib/security/ssrf.ts — refuses cloud-metadata, RFC1918, loopback, link-local hosts on every user-supplied URL fetcher.
·Cron credit-gate caches removed — each row re-checks balance so a multi-watchlist tenant can't overshoot.
·JSON parse failures in the AEO rewriter + counter-campaign drafter now save a "parse_failed" row instead of throwing after credits already burned.
·Mobile tables fixed (overflow-x-auto on /seo, /benchmarks, /agency).
·Tenant-boundary GET hardening across /aeo/runs, /counter-campaigns.
Agency dashboard + Cross-source insights
2026-05-07v0.82
·/agency — every Looper HQ tenant you have access to in one table. AEO score, competitor changes (7d), leads (30d), active sequences. 1-click switch into any client account.
·/insights — fuses AEO + competitor watchlist + SEO rank + CRM data. Surfaces moments like "X recent leads matched a query AI assistants aren't citing you for" with deep-link CTAs.
·Computed live (no precompute cron). The "act on the intel" play that pure-intel platforms can't do.
Free competitor traffic + Industry benchmarks
2026-05-07v0.81
·Free competitor traffic snapshot per domain via Cloudflare Radar + Wayback Machine — no key required, cached 24h.
·/benchmarks — anonymized industry stats (P25/P50/P75/P90/mean) for 7 metrics across all Looper tenants in your industry. Sample-size guard of 5+ before publishing.
·Daily benchmark cron computes from real tenant data — emails sent, open rate, leads, voice calls. Updates nightly.
SEO research (DataForSEO BYOK) + Similarweb passthrough
2026-05-07v0.80
·/seo — keyword research, SERP rank tracker, backlinks summary via your own DataForSEO API key. 24h cache.
·Daily SERP rank-tracker cron — track up to 100 keywords, plot the trend chart over time.
·/api/similarweb — pass-through for Similarweb visits + referrers when you connect their key.
·Both BYOK-only by design. You bring your provider account; we add the operational layer.
Competitor watchlist + Action engine + Meta Ad Library
2026-05-07v0.79
·/competitors — track up to 25 competitor URLs per watchlist. Daily diff cron detects pricing changes, new offers, copy shifts.
·On a high-significance change, the action engine auto-drafts a counter-campaign (Meta + Google ad copy + 4-step email sequence + landing hook) for human review.
·/ad-intel — pull Meta Ad Library entries for any FB advertiser. Free public source, charges only on real inserts.
·The thing pure-intel platforms can't do: see the change, ship the response in 1 click.
AEO citation tracker + Auto-rewriter
2026-05-06v0.78
·/aeo — daily scans across ChatGPT, Perplexity, Claude, Gemini, Sonar Pro, Sonar Reasoning Pro. Records whether your brand is cited for the queries that matter to you.
·Auto-rewriter proposes a Sonnet-pass rewrite of pages that aren't winning citations — clear definition, FAQ schema, named entities, no fluff.
·Daily cron, per-tenant credit gate, score history per watchlist.
·The category Similarweb just launched — without their proprietary panel, but with a closed-loop "rewrite to win" that they don't have.
Production hardening — 7 audit rounds, 60+ fixes
2026-05-06v0.71-v0.77
·Site-wide paid-action gating: every paid feature now confirms cost before charging. ~30 silent-charge bugs closed.
·Closed a credit-grant exploit where any signed-in user could RPC marketing.add_credits to gift themselves credits. Revoked EXECUTE from anon+authenticated.