> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ideatura.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Context and instructions

> Choose useful context and keep longer tasks coherent.

Use this map to choose what belongs in the current request, what is saved, and what must be checked again.

<h2 id="instructions-and-saved-work">
  1\. Instructions and saved work
</h2>

| Item | Purpose | Check before relying on it |
| - | - | - |
| Chat context | Information used in the current task | Confirm relevant sources and decisions are available. |
| Project files | Saved reference material | Identify the exact file/page; storage does not establish that it was read. |
| Guidelines | Standing workspace instructions | Confirm the workspace and test the rule in a new task. |
| Skill | A focused reusable procedure | Attach the intended Skill and inspect whether its required steps were followed. |
| Automation | Runnable work with defined inputs | Inspect the native result and reopen the correct saved version. |
| AI Flow | AI work during execution | Check coverage and credits for each run. |

Use [Guidelines and skills](/automation/knowledge) for creation steps and [AI Flows](/automation/ai-flows) for execution checks. Neither stored instructions nor workspace folders establish automatic learning or a complete history of project decisions. Keep a reviewed handover for facts that must survive a new chat.

<h2 id="choose-context">
  2\. Choose context
</h2>

<span id="choose-context-overview" />

In Ideatura, task context can come from your prompt, Guidelines, selected skills, attached files, pinned references and information read through a connected application or tool. Be explicit about which source establishes which fact.

Attach the relevant drawing page or detail, name the target view or elements and state dimensions or units. A whole project upload does not establish that every page has been read.

<span id="choose-context-in-ideatura" />

1. List the facts needed for this task.
2. Attach the relevant sources and identify their roles.
3. Specify the model, document, view and scope.
4. Ask the assistant to identify unresolved information before making a large change.

<span id="choose-context-check-the-result" />

The request contains enough evidence without burying the task in unrelated material.

<h2 id="context-window">
  3\. Context window
</h2>

<span id="context-window-overview" />

The context window is the model’s working space for a request. It is measured in tokens: pieces of text or other model-encoded information. Relevant instructions, conversation content and tool results take space; the model also needs capacity for its response and, depending on the model, reasoning.

It is different from your saved conversation, workspace storage and credit balance. A file can be stored in Projects without its entire contents being available to the model on every turn.

<span id="context-window-in-ideatura" />

1. Identify the current model and route.
2. Supply the information needed for the current task.
3. Keep the intended result and constraints explicit.
4. Use the documented capacity or indicator for that route when available.

<span id="context-window-check-the-result" />

You distinguish working context, saved material, usage tokens and credits.

<h2 id="long-conversations">
  4\. Long conversations
</h2>

<span id="long-conversations-overview" />

Long conversations accumulate instructions, revisions and tool output. When the task changes substantially, preserve a short handover: objective, current file/revision, completed changes, unresolved issues and the next bounded step.

Before continuing in a fresh conversation, reopen the saved result and attach the handover and required references. Starting a new chat does not automatically transfer the old conversation’s working context.

<span id="long-conversations-in-ideatura" />

1. Save and inspect the current result.
2. Write a short handover with the authoritative file and decisions.
3. Start a focused continuation when useful.
4. Confirm the new conversation has the needed sources and constraints.

<span id="long-conversations-check-the-result" />

The next step starts from the correct result without silently losing decisions.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.