Materialize connected text and references into a stable, reusable corpus so the same context does not need to be pasted into every prompt.
Where: Flowgen → Add → Bring in → Context
Key ideas
- Context versus Knowledge Search: Context holds the connected corpus verbatim; Knowledge Search retrieves a query-sized slice from Brain records.
- Stable append-only order: Each connected source/handle receives a durable block key. Existing blocks keep their position and newly wired sources append, rather than following incidental edge order.
- Text plus references: Flattenable sources become context text. Media or documents that cannot be flattened remain labeled references, and the combined content receives a stable context key.
- Context card: ContextNode uses the generic shared-card path. Materialize connected text and references into a stable reusable whole-corpus context block.
- Properties sidebar: contextNode.tsx
- Opened surfaces: Structured/full result preview
Steps
- Connect the text, documents, media, or source outputs that belong in one reusable corpus.
- Run Materialize context and inspect the block list, character count, and reference count.
- Connect the Context output to prompts or systems that should receive the same corpus.
- Rerun after changing sources; the context key changes so downstream caches know the material changed.
Context controls and presets
- Add path: Flowgen → Add → Inputs → Context.
- Card: ContextNode.
- Properties: contextNode.tsx.
- Controls: Connected blocks; stable append order; character/reference counts; context key; Materialize context.
- Opened surfaces: Structured/full result preview.
Tips
- Use separate Context nodes for unrelated jobs instead of building one enormous universal corpus.
- Name upstream sources clearly before materializing so block labels remain understandable.
Limitations and important notes
- Context does not semantically retrieve the best passages; it carries the corpus. Use Knowledge Search when query-time retrieval is required.
Troubleshooting
A newly connected source is missing downstream
Rerun Materialize context, then rerun the dependent node so it receives the new context key and content.