Ask anything
“How much waste did the team find this week?” “Which portfolio is most off pace?” Plain English in, account-grounded answers out — with the numbers to back them.
Flume is the team’s Generalist — ask anything about your account, generate structured reports, or plug your own AI into ROASt over MCP. Same model class as ChatGPT; the difference is the 74 ROASt tools underneath — and the guardrail that it never makes budget decisions on its own.
Flume reads your campaign data, portfolio settings and engine outputs through the same tools the agent team uses — so its answers come from your account, not from vibes.
“How much waste did the team find this week?” “Which portfolio is most off pace?” Plain English in, account-grounded answers out — with the numbers to back them.
Pre-built prompts for a Performance Review, Pacing Review, Recommendations, or Account Health Check — saved to the Documents sidebar, exportable as HTML or PDF, ready to forward to a client.
Flume can read everything and propose actions — but all budget and target recommendations come from the deterministic engine. An LLM never decides where your spend goes.
Hand the controls to your own Claude, ChatGPT or Gemini via ROASt’s Model Context Protocol endpoint — the same tools Flume uses, under the same guardrails, driven from wherever you already work.
46 read tools — campaigns, portfolios, goals, audits, reconciliation, briefs, keyword + creative analytics — plus 28 write tools covering budgets, goals, events, automation and pause applies. Mint a token in Settings → API Access, paste it into your MCP config, done.
Writes route through the same staging + audit pipeline as the in-app Optimiser. Destructive deletes require an explicit confirm flag — refused otherwise. Your AI proposes; ROASt enforces.
Every token is minted with a capability scope — read-only, write, or full — and can be revoked at any time. Give a reporting bot read access without ever exposing a write surface.