Glossary of Terms

• Feeda Studios — Physical and/or virtual AI development studios modeled on music studios. Each studio is a standardized environment—hardware, software, roles, and processes—to build AI products (agents, copilots, multi-agent networks).

• Studio Session — A time-boxed, structured development engagement (e.g., half-day, full-day, multi-day) where a product owner and a Feeda Studio team co-create or enhance AI products.

• AI Studio Engineer — The session’s technical lead (analogous to a music studio engineer). Responsible for the workstation, model stack, orchestration tools, evaluation harnesses, safety rails, and release “mastering.”

• Domain Expert / Context Engineer — The subject-matter authority for the product’s vertical (e.g., automotive, hospitality, healthcare). Curates and structures context packs (documents, datasets, procedures) and practices context engineering: the evolution of prompt-engineering that organizes knowledge, guardrails, workflows, and retrieval.

• Prompt Engineer — Crafts optimized dialogue flows, system prompts, and tool-use scripts for models and agents. Works closely with the context engineer and product developer.

• Product Developer — Focuses on the UI/UX, APIs, and integrations; connects frontend surfaces (web, mobile, tablet) to backends (agent runtimes, vector stores, model gateways).

• Product Owner / Rights Holder — The company, individual, or team funding and hosting the session. Owns the outputs (unless negotiated otherwise).

• Master Agent — The canonical, “gold-stamped” AI product artifact (agent, copilot, or multi-agent network) produced from a session. Analogous to a master recording.

• Stems — Composable assets that contribute to the Master Agent (e.g., prompts, tools, knowledge packs, finetuned model weights, vector indices). Analogous to stem tracks in audio production.

• Tokenization — Converting ownership of the Master Agent (and optionally stems) into transferrable digital tokens, enabling licensing, resale, and royalty flows. (See: Feeda 10k GitBook & Feeda Litepaper.)

• Royalty Splits — Negotiated percentages for ongoing commerce on an agent (e.g., per-seat subscription, API usage, resale, marketplace licensing). Enforced by smart contracts compatible with standards like ERC-2981 (royalties), ERC-721 (non-fungible master), ERC-1155 (multi-edition), and optional ERC-4907 (rentals).

• Context Pack — Curated corpora (docs, policies, data dictionaries, SOPs) transformed into AI-ready retrieval units (chunked, embedded, versioned).

• Agent Orchestrator — Framework that coordinates tool-use, planning, and multi-agent workflows (e.g., LangGraph-style DAGs, Crew-like role agents, AutoGen-like conversational loops).

• RAG (Retrieval-Augmented Generation) — Pattern where models cite and reason over external sources; the studio normalizes this into Context-First RAG (structured source-of-truth retrieval precedes generation).

• Evaluation Harness — Metrics and tests (functional, safety, bias, grounding, cost, latency) operated as part of the session; produces objective “mastering” gate checks.