Difference between, context, knowledge and integration

In Goodweek, when you chat with the AI, the quality of the answers depends largely on the information made available to it. To answer more accurately and precisely, the system can rely on four major “sources”: instructions, context attachments, space knowledge, and integrations.

These mechanisms can sometimes look similar in intent (providing more context), but they don’t have the same role or the same effects. Distinguishing them helps explain why Goodweek offers several paths rather than a single one.

Instructions

Instructions act as a “framework”: they guide how the AI should interpret your request and formulate its response. They are especially useful when you need to make explicit information that would otherwise remain implicit (project stakeholders, goals, constraints, internal vocabulary, expected tone, desired structure, etc.).

They can also be used to calibrate style and expectations (example answers, formats, sections), in a way that’s close to few-shot learning: rather than describing at length what you want, you show what a good answer looks like.

Context attachments

Attachments address a different need: providing “raw material” (a document, a note, an export, a PDF…) that the model can read and use to produce an answer or generate new content while staying faithful to the substance (and sometimes the format) of the provided document.

This approach is very effective when you need to be faithful to a specific source. On the other hand, adding too many attachments can saturate the available context and degrade quality (up to incomplete answers or hallucinations): this isn’t a problem of intent, but a capacity limit. That’s why Goodweek optimizes how files are processed (for example, via “compressed” versions used in chunks), to balance cost and quality.

Space knowledge

Space knowledge exists for another use case: having a stable, durable, reusable foundation without having to attach the same documents to every message. Instead of loading an entire file each time, Goodweek can search for relevant passages in the space’s synchronized documents, and then inject only what truly helps answer.

In practice, this creates an interesting trade-off between accuracy and frugality: you keep a rich documentation base while avoiding “polluting” the reasoning with full pages when only one sentence is useful.

Integrations

Integrations finally solve a problem that neither instructions nor documents handle well: accessing dynamic information, scattered outside Goodweek and difficult to maintain in a static file. In these situations, asking the AI to fetch the information where it lives (emails, CRM, internal tools, etc.) makes it possible to work with up-to-date data.

In other words, an integration doesn’t replace a knowledge base; it temporarily broadens the information scope for a given question.

In summary

• If you need to provide framing information (goals, constraints, expected style), instructions are the right lever.

• If you need to be faithful to a specific document, prefer a context attachment, while keeping the risk of overload in mind.

• If you have durable, shared documentation, use space knowledge, which searches and extracts only what’s relevant.

• If the information is “alive” and already lives elsewhere, integrations are the most natural way to access it at the right time.