Conversational AI and anomaly analysis with visible CRM keyword search. Controlled CRM semantic assistance is off by default and limited to approved agency-assistant contexts.
Meet governed assistantsAsk questions about your clients, campaigns, budgets, and workflows in plain language. Mention any entity with @client, @task, or @project to pin context directly into the conversation.
The AI understands your agency's structure -- it retrieves relevant tasks, briefs, spend data, and timesheets to give you answers grounded in real data, not hallucinations.
Ultra-fast inference powered by Groq. Sub-second responses for real-time conversational workflows.
Type @client, @task, or @project to pin entities into context. The AI fetches their full data automatically.
Intent classification routes queries to the right data sources -- spend, timesheets, tasks, briefs, or knowledge base.
The assistant calls dedicated tools to pull live numbers on demand -- finance, ad-spend, anomalies, projects, clients, social -- and proposes any change (like creating a task) for you to confirm. It never acts on its own, and only ever within your permissions.
Switch between Finance, Marketing, Sales, and Account Management assistants -- same engine, each leading with the right focus and tools for the role.
Pin important conversations for quick access. Up to 25 pinned threads per user with dedicated sidebar section.
Ask pricing questions in plain language through authorized live data lookups. CRM semantic assistance does not broaden access to rate-card data.
Ask "How many hours on Client X last month?" and get instant answers. Timesheet gap analysis spots missing entries automatically.
A proactive agent that continuously scans your agency data and flags issues before they become problems.
Detects unusual spikes or drops in ad spend across Meta and Google campaigns.
Flags campaigns approaching or exceeding their monthly budget limits.
Identifies tasks and projects at risk of missing their due dates.
Monitors client engagement signals and flags accounts needing attention.
Tracks revenue patterns and alerts on declining or stagnant client revenue.
Analyzes team capacity and flags over- or under-utilized team members.
Evaluates campaign KPIs against benchmarks and flags underperformers.
Finds unbilled work, missing invoices, and revenue leakage across clients.
Visible agency and portal CRM search remains keyword-ranked. Semantic assistance through Cloudflare Workers AI and Vectorize is available only to approved agency-assistant contexts, covers people, companies, and opportunities, and is off by default.
When controlled assist is enabled, current authorization is rechecked and only records at confirmed index state can contribute semantic candidates. Portal semantic ranking remains unavailable.
For approved CRM assist only, Vectorize candidates are filtered by current authorization before fusion with keyword results.
Exponential decay scoring with configurable half-life per intent type.
Weighted scoring by entity type -- spend data ranks higher than board metadata.
Edge-classified intent aligns results to what you are actually asking about.
Results mentioning the same clients, projects, or tasks as your query rank higher.
AI assists at every stage of brief creation — from suggesting field values as you type, to scoring brief quality in real time, to generating complete briefs from a single paragraph.
The quality scoring engine rates each brief across three dimensions: required field completion (40%), optional fields (20%), and content quality (40%). Issues like vague language, conflicting requirements, and unrealistic timelines are flagged with specific improvement suggestions.
AI suggests values for empty fields as you type — target audiences, platforms, timelines — based on context and past briefs.
Provide a one-paragraph description and AI populates every template field. Accept, edit, or reject each suggestion individually.
AI reviews completed briefs for vague language, conflicting requirements, and unrealistic timelines with specific fix suggestions.
Powered by Cloudflare Workers AI, intent classification and summarization run at the edge for sub-50ms latency with Groq fallback.
Queries are classified at the edge into financial, process, search, time tracking, pricing, and other intents — routing to the right data sources in under 50ms.
Long documents and thread summaries are generated at the edge via Llama 3.1 8B. No round-trip to external APIs — everything stays on Cloudflare's network.
If Workers AI is unavailable, requests automatically fall back to Groq for classification and generation. Zero downtime for AI features.
Train custom LoRA adapters on your agency's chat history, knowledge base, and workflows. Route traffic between adapters with weighted A/B testing.
Five extractors pull chat Q&A pairs, intent examples, RAG documents, knowledge entries, and combined datasets. Automatic anonymization and JSONL export to R2.
Upload, activate, and retire LoRA adapters. Weighted traffic routing lets you A/B test new models against the baseline with gradual rollout.
Track adapter performance across response quality, latency, and user satisfaction. Compare adapters side-by-side and promote winners automatically.
Stop digging through dashboards. Ask the AI, and it finds the answer.
Get Started