How the AI chat works

What is AI context

AI context is the set of information you attach to a conversation so the AI can answer with the right details. It can include documents, short excerpts from documents, and notes you created in the workspace.

Here is how it works in the UI:

1. Add items from search results using Add to context.
2. Use the attachment menu in the chat input to pick documents or notes directly.
3. Review the context panel above the chat box to see what is included.
4. Remove anything that is no longer relevant before you ask your question.

The AI does not scan your whole workspace by default. It only reasons from the items you added, which keeps answers focused and easier to verify.

Want to find the right documents first? See Search in Contradic.

Global search

You can change how the AI sources its information using the Sources toggle near the context panel:

• Selected sources: The AI will only provide answers from the documents and notes you explicitly added to the context. This keeps the answers highly targeted and focused.
• Search all sources: The assistant will automatically search across all available knowledge sources in your workspace to find the best context to answer your question. This is ideal when you don’t know exactly which documents contain the answer.

Want to understand how this search pulls the best evidence without reading your entire workspace? See How Global Search works.

Note: Adding extracts directly to the context from the chat screen requires you to be in the Selected sources mode.

Leave and return while Contradic responds

After you send a message, you can open another conversation, visit another page, refresh the browser, or close the page. Contradic can continue generating the response in the background. Navigating away does not cancel the request.

To return to the response:

1. Open Manage conversations.
2. Select the conversation where you sent the message. A rotating arrow may identify a conversation that is still generating during your current session.
3. Keep the conversation open while Contradic checks its status. The saved response appears when it is ready, although the earlier live progress is not replayed.

If you see a message that the assistant is still generating in the background, reopen or refresh the conversation shortly.

To cancel the request instead, select Stop generation in Selected sources mode or Stop search in Global Search mode before leaving the conversation.

Context limits

Your chat has a capacity for context that you can adjust depending on your needs. Use the Context size toggle to select between:

• Standard: Capped at 270,000 tokens (roughly 500 pages). This is the default and is required if you are using the Search all sources mode.
• Max: Uses the full context limit of the selected AI model (e.g., up to ~800,000 tokens, or roughly 1500 pages, on advanced models). This allows for much larger collections but is only available when using Selected sources.

You can see this capacity in the context meter on the right of the chat input:

• The percentage shows how full the context is based on your selected size.
• When it turns red, the context is over the limit.
• If the limit is reached, you will not be able to add more items until you remove some.

To stay within the limit:

1. Attach only the most relevant documents or excerpts.
2. Remove older items that are no longer needed.
3. Start a new conversation if you want to switch to a different topic.

Rate an AI answer

After an assistant answer finishes, use its action bar to tell Contradic whether it helped:

1. Select Helpful or Not helpful below the answer.
2. Select any reasons that apply. Helpful feedback offers Helpful, Accurate, Clear, and Complete. Not helpful feedback offers Inaccurate, Incomplete, Unhelpful, Irrelevant, and Other.
3. Add an optional comment of up to 500 characters.
4. Select Submit feedback.

The selected thumb appears filled after your feedback is saved. Feedback controls are available for completed answers in saved conversations, not while an answer is still being generated.

To change your feedback, select the other thumb, adjust the reasons or comment, and submit again. The new feedback replaces the previous response for that answer.

To remove your feedback entirely, select the currently active thumb. This clears its rating, reasons, and comment.