claude-agents8 min
Claude AI agents
Roles, prompts, escalation paths, audit trail.
Four agents · one warehouse
Each Traqly workspace ships with four autonomous agents (Scaling, Creative Fatigue, Attribution Stitching, Predictive Bidding). All four read your server-side warehouse via tool-use. They never see PII — only aggregated KPIs and structural metadata.
BYO prompt
Scale plans can override the system prompt for any agent. The runtime executes the prompt with a curated tool set (campaigns.query, creatives.query, attribution.query) and your customizations.
# .traqly/agents/scaling.md
You are an aggressive media buyer. Optimize for blended ROAS > 4.0 with a daily change budget of $500.
Always justify your recommendation with a one-sentence cause from the underlying data.Audit trail
Every agent decision writes a row to AgentDecision with the input query, the tool calls, and the suggested action. You approve before any change ships to Meta/Google/TikTok/YouTube APIs.