Glossary
The terms used across the product. They stay in English in every language, so teams share one vocabulary.
Updated September 25, 2026 · 2 min read
Terms
| Term | Meaning |
|---|---|
| Canvas | The visual editor where the workflow is drawn. |
| Root Orchestrator | The workflow's entry point: global behavior rules, analytical engine and token budget. One per workflow. |
| Sub-Agent | A specialist agent the Root (or another Sub-Agent) delegates to. |
| Skill | A tool an agent can call, defined by a function name, a description and a JSON Schema. |
| Reference | Context for an agent: a document, data or a URL, with reading guidelines. |
| Connector | Access to an external tool through an MCP server and a credential kept in the Vault. |
| Pre-Flight | The live validation panel: every problem that would block an export or a run. |
| Blueprint | The files compiled for a target environment (for Claude Code: CLAUDE.md, Skills, .mcp.json). |
| AST | The workflow as structured data (JSON): what the Canvas edits, what gets exported and run. |
| CLAUDE.md | The instructions file Claude Code reads at the root of a project. |
| MCP | Model Context Protocol: the open standard agents use to call external tools. |
| OAuth | Sign-in on the vendor's site that gives the app a token, without sharing a password. |
| BYOK | Bring Your Own Key: Live runs use your organization's Claude API key. |
| Extended Thinking | Reasoning the agent's brain does before it answers, with an effort level. |
| Artifact | A self-contained output (document, code, table) the console renders on its own. |
| SSE | Server-Sent Events: how the console receives a run's events live. |
| Payload | The data passed along a link or a tool call. |
| Webhook | A call a service makes to the app when something happens (e.g. a payment). |
Business vocabulary
The interface speaks the language of the business. For technical readers, this is what each name corresponds to, and how it works in this product:
| In the product | Technical term | Here |
|---|---|---|
| Behavior rules | System prompt | The instructions an agent follows on every run: the Root's global rules, each Sub-Agent's own. |
| Agent brain (analytical engine) | Model, LLM | The Claude model that powers an agent: Claude Opus 5 unless a Sub-Agent sets its own. |
| Company Memory | Knowledge base, vector store, embeddings | References: documents and data given to the agents as context. There's no vector store and no embeddings. |
| Document reader | Text splitter, reading instructions | How an agent reads a Reference (tables, visuals, structure, code). Documents are read whole: they aren't split into chunks. |
| Database lookup | RAG pipeline | Querying company data through a Skill connected to your database (Postgres, MongoDB, Supabase): a tool call, not a retrieval pipeline. |
Why these stay in English
Architecture words drift when each team translates them its own way. Keeping them identical in every language means a Portuguese and an English team can read the same Canvas, the same Blueprint and the same docs without a translation table.