KnaveAI
Product

Twelve surfaces.
One brand voice.

Knave is a marketing function, not a tool. Every surface ties back to one brand profile, one approval queue, one voice judge. Below is what it actually does.

01

Brand profile review

Knave reads your site, PDFs, and Google Drive. Then shows you what it learned and asks you to correct it. Approval locks the profile as ground truth for everything else.

Site crawl up to 200 pages tier-gated, chunked at 512 tokens, embedded with text-embedding-3-large, stored in pgvector with hnsw. Site + PDF + Google Drive at v1. Notion + Dropbox + ZIP + Slack at v2.

Spec
  • ▪Pages crawled per re-index: 50-200 (tier)
  • ▪Embedding model: text-embedding-3-large
  • ▪Re-crawl cadence: weekly + on-demand
  • ▪Cost per re-index: ~£0.03 (incremental)
02

Content engine

Blog posts, social posts (IG, FB, LinkedIn, X), email campaigns, ad creative, landing pages. Per-vertical templates. Approval queue. Autopilot per-surface after 20 customer approvals.

Quality judge gate on every output. Voice match scored 0-100 against your brand profile. Brand consistency judge on images. Cross-asset coherence gate when a campaign produces blog + social + email together.

Spec
  • ▪Blog draft: ~£0.03
  • ▪Social draft: ~£0.01
  • ▪Email draft: ~£0.02
  • ▪Image: ~£0.06 (gpt-image-2 high quality)
03

Active social engagement

Read every comment on your IG and Facebook posts. Knave drafts a reply in your brand voice within 5 minutes. You approve in seconds. After 20 approvals, optionally let Knave handle similar comments on autopilot.

LinkedIn monitor-only at v1 (auto-reply at v1.5 once detection-risk is modelled). X and TikTok at v2. Regulated verticals (financial / legal / medical / healthcare) get autopilot permanently disabled. Drafts only.

Spec
  • ▪Ingest latency: < 5 min
  • ▪Reply latency target: < 2h post-comment
  • ▪Autopilot threshold: 20 approvals per platform
  • ▪Per-customer cost at 100 comments/mo: ~£0.45
04

Site audit

Weekly Playwright + Lighthouse run. Tech-SEO issues, content gaps, schema gaps, Core Web Vitals regressions. One-click 'Have Knave fix' button auto-drafts the correction into your queue.

DataForSEO integration for keyword intel. Cross-references GSC ranking data for fix prioritisation. Output severity-ranked: Critical (blocks indexing), High (rank-loss risk), Medium (optimisation opportunity).

Spec
  • ▪Cadence: monthly Solo / weekly Growth / daily on key pages Plus
  • ▪DataForSEO cost: ~£2-5/customer/mo
  • ▪Findings retained 12 months
05

Competitor monitoring

Add 3 competitor URLs. Knave runs a weekly Playwright crawl, hash-diffs change, summarises moves with Haiku. Daily competitor brief in your morning digest.

Pricing-page changes flagged immediately. Review-score deltas via SerpAPI google_local. Public social-post velocity for IG/FB. Used by the proactive prompts engine to suggest counter-content.

Spec
  • ▪Per-customer cost: ~£0.30/mo at scale
  • ▪Competitors per tier: 3 / 3 / 10 / unlimited
  • ▪Detection: pricing, services, blog, GBP reviews, public social
06

Multi-engine analytics

GA4 + Google Search Console at v1. Bing Webmaster Tools and Cloudflare Analytics at v1.5. One adapter pattern, one dashboard, one Insights inbox.

Reports tab cross-references GSC ranking position with DataForSEO keyword intel: 'you rank 7 for X, top page is 1,400 words, yours is 500. Want a rewrite?' All free APIs. £0 in adapter cost beyond OAuth maintenance.

Spec
  • ▪Latency: 2-3 days behind real (GA4 + GSC limits)
  • ▪Adapter pattern at lib/analytics/
  • ▪Free APIs, customer-OAuth-scoped
07

CRM-lite

Default contact database for customers without Klaviyo or Mailchimp. Lead capture from chat widget, newsletter form, manual entry, CSV import. Lifecycle automations (welcome / re-engagement / win-back). Send via Resend on your verified domain.

When you connect Klaviyo later, contacts dual-write or migrate one-way. Your call. Double-opt-in at capture. Unsubscribe one-click. GDPR data export and deletion built-in.

Spec
  • ▪Per-customer cost: ~£0.60/mo (Resend + storage)
  • ▪Default sender: Resend on customer-verified domain
  • ▪Migration: one-way push or dual-write to Klaviyo / Mailchimp / Brevo
08

Chat widget

One JS snippet on your site. Brand-RAG-grounded answers. Captures leads to CRM-lite. Vertical-specific never-answer rules so regulated businesses don't hand out advice.

30-day human review per install before chat autopilot unlocks. Hard cap on Claude Haiku token spend per installation. Email-prompt after 3 turns. 20-turn session memory, localStorage-keyed.

Spec
  • ▪Quota: 200 / 1k / 5k msg per month per tier
  • ▪Backed by Claude Haiku for low latency
  • ▪Lead-capture trigger: turn 3+ with intent signal
09

Knave's suggestions

Each morning Knave deposits 3-5 specific suggestions in your inbox: industry news angles, trending topics, holiday tie-ins, competitor moves, performance anomalies. One click drafts the content.

Engine pulls NewsData.io + Inoreader RSS + SerpAPI Trends + X API + Calendarific. Shared infrastructure across all customers. Per-vertical relevance scored by Haiku classifier.

Spec
  • ▪Shared infra: ~£100/mo across all customers
  • ▪Per-customer cost: ~£2-5/mo at 50 customers
  • ▪Cadence: 7am customer timezone
10

Automated insights

Eight categories of internal-data signals delivered to dashboard, daily email digest, and iOS push: rank opportunities, content gaps, performance anomalies, audit findings, engagement spikes, competitor moves, lead-source shifts, voice-match drift.

Hard 3-push-per-day cap with quiet hours. No empty-state emails (don't manufacture content for cadence's sake). Outcome tracking: 7 days post-action, did the predicted impact materialise?

Spec
  • ▪Categories: 8
  • ▪Push cap: 3/day, quiet hours 9pm-7am
  • ▪Outcome tracking: 7-day post-action sample
11

Voice judge

Sonnet-scored voice match on every text output, 0-100. Below 70 surfaces with a low-confidence badge. Below 50 regenerates. ≥85 required for autopilot send. Calibrated against your approved content over time.

Brand consistency judge on images runs on a 0-5 scale: palette match, brand mark presence, composition. Below 3 regenerates once and flags. Generic AI writing structurally cannot reach the autopilot threshold.

Spec
  • ▪Text scale: 0-100 (Claude Sonnet rubric)
  • ▪Image scale: 0-5
  • ▪Autopilot floor: 85 voice / 4 image
12

Setup with achievements

Customers reach 100% configured through 18 weighted tasks across Knowledge, Connections, Content, and Polish. 13 named achievements (First Move, Brand Crafted, Voice Found, Autopilot Ready, 100% Set Up). Linear-tone restraint, no Duolingo guilt.

Setup % at day 14 is our churn predictor. Tier-aware scoring (Solo's 100% excludes custom-domain task). Internal-dogfood path pre-completes imported achievements so the 5 founder-portfolio installs don't redo prior setup.

Spec
  • ▪Tasks: 18 (1 tier-gated)
  • ▪Achievements: 13 (3 rare, 4 uncommon, 6 common)
  • ▪Daily Driver = 7 consecutive approvals (only streak)

See it run on your business.

Start 14-day trial