An agent orchestrating a ten-step sequence spends a turn per step and can lose the thread halfway. Sending a typed plan that the server executes — with real branching and iteration — makes the sequence deterministic and cheap. It is the same instinct as CorpusIQ's runbooks, expressed as a language rather than a catalogue. Whether synthetic audiences answer your question is a separate judgement, and one worth making before designing around the output.
Neuroflash's endpoint for their synthetic-audience product. What distinguishes it is Plan-Mode: instead of the client orchestrating a sequence of calls, a typed plan with branching and loops runs server-side.
- twins.query — poll synthetic audiences and segments
- content.generate — branded content generation
- Three modes: traditional per-endpoint tools, a server-side typed Plan-Mode supporting branching and loops, and an exploratory beta mode
- Around 88 tools in total
OAuth 2.0 with PKCE, handled by the client. One neuroflash account covers the app, the API and MCP.
Paste a URL, then authorize — add the endpoint to your client, then approve the OAuth consent screen