Conversation & Input

Attachments

Drag files to the composer or click the paperclip:

50MB size limit per attachment.

Slash Commands

Type / for the command panel: force-invoke skills, switch modes, or enter plan mode.

@ References

Type @ to reference any of four kinds of objects and let the agent work directly with them:

• Files: artifacts from this session or files in the workspace; type part of the name to find them
• Sessions: past sessions in the current project, bringing earlier conclusions into context
• Board tasks: reference by identifiers like LAN-1; the agent reads the task details before acting
• Experts: bring a specific expert into this turn—handy when you want a particular perspective

You can reference several at once; each shows as a removable chip in your message. It covers “fix it per this task”, “continue from last week’s report”, “have legal take a look”—no copy-pasting back and forth.

Text Quotes and Annotations

Select any text in a LanMate reply to get the floating menu:

• Quote to composer: quotes with the original text and position markers; multiple quotes can be active at once and reordered, so follow-up questions focus on exactly those passages
• Add comment: inline annotations on the selection, handy for flagging “this is wrong” or “change here”

Quotes can be edited and removed later, with original-text highlighting preserved—especially useful for pinning down discussion points in long replies.

Paste Behavior

Everything pasted into the composer is normalized to plain text (rich-text formatting copied from web pages or AI replies won’t linger). Image and file pastes, and the auto-conversion of very long text into attachments, work as before.

Turn Modes

The mode label above the input controls how the current turn runs. Besides “Auto”, three controlled modes share one idea: think before you act.

| Mode | In one sentence | What the agent can do | |—|—| | Auto (default) | Just do it | All tools | | Plan | Propose a plan, act after you approve | Read-only (read files, inspect code, ask clarifying questions); nothing gets modified | | Spec | Write a spec first, confirm acceptance criteria, then implement | Draft phase is read-only; full tools return after you confirm, executing per checklist | | Brainstorm | Talk ideas only, no action | Read-only + web research + expert consultation; no file changes, no commands |

Plan Mode

For tasks where the approach is contested and you want alignment before work starts. The agent studies the current state and proposes a step-by-step plan; execution begins after you approve, or just say “make step 2 write the tests first” to adjust.

Refactor the user module; give me a plan first

Spec Mode (coding tasks)

For coding tasks with complex requirements that need explicit acceptance criteria. The agent produces three documents (spec / task list / acceptance checklist); after you confirm in the UI, it implements per the list, checking off items as it goes and closing automatically when all are done.

Use Spec mode to implement Excel export

The confirmation step is a gate: approve directly, or send it back with revision notes. The spec stays in the project for future iterations against the same standard.

Brainstorm

For the stage where you don’t yet know what you want. The agent explores angles, asks questions, and plays devil’s advocate to sharpen a vague idea—without touching files or running commands. Once the idea is clear, switch back to Auto and let the agent build it.

Brainstorm: who is this dashboard for, and which metrics should it show?

Switching

• The mode label above the input, switchable anytime, affects only the current turn
• Slash commands: /plan, /spec, /brainstorm
• Spec returns to Auto automatically when finished

Context Usage

After each turn, the session header shows current context usage (a bar plus percentage). When it fills up you can act proactively instead of waiting for auto-compaction:

• Keep digging: over 80% usage, run /compact to compress manually—key conclusions stay, process details go
• Change direction: start a new task; context starts fresh
• Quick question: switch to a side chat and leave the main session’s context untouched

Auto-compaction triggers near the window limit and leaves a “Context Compacted” marker. Compaction loses detail—important conclusions, file paths, and data definitions are best saved to notes or files before compressing.

Drafts

Per-session drafts auto-saves locally. Preserved across session switches and app restarts. Cleared on session archive.

Goal Mode

For long tasks, enable Goal Mode: set a clear target, and the agent works step-by-step with progress reports until completion.