Chat
Chat is Feldar’s project-aware conversational assistant. Use it to brainstorm, examine a passage, plan revisions, or propose changes while keeping control over what reaches your project.
This page describes the complete desktop Chat workspace. The native iOS app provides an adaptive Chat surface for project conversations and editor proposals, including sessions, attachments, and review-before-apply proposal cards. Lens personas and the visual Canvas workspace remain desktop workflows. See Chat and Rewrite in the mobile app. Chat is not part of the streamlined mobile-web workspace.
What you can do
Section titled “What you can do”- Brainstorm story ideas and plot directions
- Ask questions about the current page and project context
- Compare approaches to a scene or revision
- Get feedback through Lens personas and Reader Reactions
- Review proposed editor changes, project-wide replacements, and Canvas cards before applying them
How it works
Section titled “How it works”- Open Chat from the editor header
- Choose a model from the composer
- Add relevant context and write your request
- Review the response
- Apply only the suggestions or proposals you want
Sessions
Section titled “Sessions”Chat keeps separate sessions for a project. Choose the session title in the Chat header to switch conversations, or choose Create new chat to start with an empty thread. Use separate sessions for unrelated goals so their context stays easier to follow.
While a response is streaming, stay in the current session. Switching sessions and creating a new chat become available again after the response finishes.
Panel layouts
Section titled “Panel layouts”On desktop, Chat can use either of these layouts:
- Side panel — docked beside the editor and resizable
- Floating panel — an overlay that can be moved around the workspace
The compact instruction box that appears after selecting text belongs to the editor’s Instruct rewrite flow. It is not a third Chat mode.
Models and reasoning
Section titled “Models and reasoning”Choose a model from the lower-left side of the composer. Model availability depends on the current catalog and your plan. When a selected model supports adjustable reasoning, its menu also lets you choose the reasoning effort. Models without supported reasoning levels do not show that control.
Attaching context
Section titled “Attaching context”Use the attach button to add images or supported document files. You can also drag files, text, or a selection from the editor into Chat. Attached items appear above the composer and can be removed before sending.
For passage-specific questions, select or drag only the text that matters and explain what you want assessed. A focused request usually produces a more useful answer than attaching unrelated material.
Image analysis depends on the selected model. If a model cannot accept images, Feldar rejects the request instead of silently ignoring the attachment. Remove the image or choose another available model that supports images, then send the request again.
Long conversations and manuscript reviews
Section titled “Long conversations and manuscript reviews”Chat keeps recent turns in detail and condenses older parts of a long conversation so the session can continue. Important decisions remain useful as context, but exact wording from much earlier turns may be shortened. Restate an important constraint or attach the relevant passage again when precision matters.
A large manuscript review may span more than one turn. Keep using the same session and ask Chat to continue with the remaining pages when it says more material remains. If the review scope matters, ask which pages it covered.
Web research
Section titled “Web research”Web research is deployment-dependent and may not be available in every Feldar environment. When it is enabled, Chat can automatically search the public web for current or real-world information that the manuscript and Canvas cannot answer. Answers that use public sources should include those sources as links.
If current information matters, ask for linked sources. Do not assume Chat searched the web merely because a question could have benefited from it.
Lens feedback
Section titled “Lens feedback”Type @ in the composer and choose one or more personas to enter Lens mode. You can choose Normal, Debate, Synthesize, Debate & Synthesize, or Reader reactions as the response mode. See Lens for the full workflow.
Proposals and approval
Section titled “Proposals and approval”Chat can return structured proposals instead of silently changing project data:
- Proposed edit — insert or replace text on a specific page
- Proposed replacement — replace matching text across affected project pages
- Proposed card or Proposed card update — create or update a Canvas card
Review the target, proposed content, affected pages, and rationale shown on the card. Choose Apply, Insert, or Save card to approve the change, or Dismiss to leave the project unchanged. An editor proposal can be applied only while its target page is the current page.
Editor proposals pass through prose-quality checks before they reach the card. Chat first tries to revise a flagged draft. If warnings remain, the card lists them under Prose lint. These warnings are advisory and do not block Apply; they describe the proposed text and are not a judgment about whether your own writing is AI-generated.
After you apply or dismiss a proposal, Chat keeps that decision as session context for later turns. The event is not shown as a synthetic message in the visible conversation, but follow-up responses can account for whether the proposed change was accepted.
- Be specific. “Make this dialogue feel more guarded” works better than “improve this.”
- Keep one goal per session when possible
- Attach the smallest useful amount of context
- Ask follow-ups to test or refine a direction
- Read every proposal before applying it; Chat responses and proposals are suggestions