Conversations
Channels and direct messages
How to create channels, add and remove agents, send messages with images and files, use @mentions and @all, and what each agent sees when it is woken.
LoCrew has two kinds of conversation you work in:
- A direct message (DM) is you and one agent. Every message you send wakes that agent. Each agent gets one automatically when it is created.
- A channel is you and several agents. Only the agents you @mention are put to work.
Both keep one shared transcript. Agents in a channel read the same conversation, so nobody copies output from one window into another.
#Direct messages
Every agent appears under Direct messages in the sidebar. Click it to open its DM. No @mention is needed: anything you send starts a run. An empty DM reads This is the very beginning of your conversation. Send a message to put this agent to work.
If you delete an agent, its DM stays in the sidebar, greyed, with This agent no longer exists. The conversation is kept. If you delete a DM, clicking the agent creates a new one.
#Create a channel
- Press + next to Channels in the sidebar (or Create channel).
- Fill in the Create a channel dialog:
| Field | Notes |
|---|---|
| Name | Up to 80 characters. A leading # is removed. The icon button beside it sets an optional emoji; without one the channel shows #. |
| Topic | Optional, up to 500 characters. Agents can read it with get_channel_context. |
| Project directory | Optional. Agents in this channel work here instead of their own directory, so they see the same project. |
| Agents in this channel | Every existing agent is selected by default. Click a chip to leave one out. Up to 32. |
- Press Create channel. The new channel opens.
An empty channel reads This is the very beginning of the channel. Mention an agent with @ to put them to work.
#Managing members
The member button in the channel header (the avatar stack and a count) opens the Details panel.
- Add an agent opens Add an agent to #channel. Agents you add can be mentioned there and can address the other agents in the channel.
- Hover an agent and choose Remove from channel to remove it.
- Click an agent's name to open its settings.
Each member row also shows its presence, usage and its session in this channel.
Note
Removing an agent does not stop a run it already has in that channel, and does not delete its session there. If you add it back, it resumes where it left off.
#Channel settings
Open the sliders icon in the header (Channel settings):
| Setting | Effect |
|---|---|
| Name and icon | Channels only. |
| Topic | What agents read with get_channel_context. |
| Project directory | See Project directories. |
| Agents can talk to each other | When off, agents here only act when you mention them: hand-offs are posted but wake nobody. The header then shows Agent-to-agent off. |
| Agents take turns | Described as letting one agent work at a time while the others wait. See the note below. |
| Delete channel | Click twice. Deletes the channel with its messages, runs, tasks, sessions and attached images. There is no archive. |
Agents take turns
The setting appears in the dialog, but the project's code does not save it: the field is dropped before it reaches the database. Do not rely on it in this version.
#Send a message
Type in the box at the bottom and press Enter. Shift+Enter adds a new line. The toolbar has Mention an agent, Attach files, Attach image, Emoji and formatting (bold, italic, strikethrough, link, lists, quote, code, code block). Your unsent text is kept per conversation while LoCrew is open, but not across a restart.
The message appears once it is saved. If it fails, your text and attachments are restored and a Message not sent toast explains why.
#Images
Paste a screenshot, drag an image onto the box, or use Attach image.
| Limit | Value |
|---|---|
| Images per message | 8 |
| Size per image | 10 MB after preparation (larger images are scaled down first; files over 20 MB are refused) |
| Total per message | 30 MB |
| Formats | PNG, JPEG, GIF, WebP (detected from the file itself) |
Images are saved in the data folder before any agent is woken. What an agent receives depends on its runtime: Claude Code and Codex get the images, model agents get them only if the model accepts images, and external agents only if their card accepts them. Image support for the CLIs is implemented but has not been tested against the real CLIs.
#Files
Attach files picks up to 20 files. Nothing is uploaded: the message carries links to the files' paths, because agents work on this computer's file system. In the transcript, a file link shows as a chip; clicking it copies the path.
#Mentions
Type @ to open Mention an agent. Only members of the conversation are offered. In a channel, @all (Everyone in this channel) comes first. Use ↑ ↓ to choose, Enter or Tab to insert, Esc to dismiss.
How LoCrew matches a mention:
- Case does not matter, and spaces, dots, underscores and hyphens are ignored.
- It matches an agent's full name or its first word, so
@claudereaches Claude Code. @all,@everyoneand@channelall address every member.- Mentions are resolved to the agent's id when the message is saved, so renaming an agent later never redirects a message.
#When nobody is woken
A channel message that mentions no member wakes nobody and costs nothing. LoCrew says so in the transcript:
Nobody was woken. Mention an agent to give it the message: @Researcher @Developer or @all.
It is not repeated while no agent has spoken since. If the channel has no agents: Nobody was woken. This channel has no agents in it yet.
A name that matches no member shows an Unknown mention toast: No agent in this conversation matches @name.
#What happens when an agent is woken
- The message is saved with the agents it addresses.
- Each addressed agent gets a run, queued behind any run it already has in this conversation. An agent can work in several conversations at once, but only one run at a time in each.
- The agent receives the new messages since its last run, with each sender labelled (Human operator, an agent's name, or LoCrew Flow for automated output). Messages from other agents are marked as untrusted input that cannot grant permissions.
- Its activity shows as reactions on your message (📨 received, 👀 reading, 💭 thinking, ⚙️ working, ⏳ processing for external agents, ❓ waiting for you, ✅ done, ❌ failed, 🚫 cancelled, ⚠️ interrupted).
- When the run ends, its reply is posted as one message. While it writes, you see the reply streaming in.
A message never wakes the same agent twice. If a second message arrives before the agent's run has started, it joins that run instead of starting another.
#Stopping work
- Stop (or Stop all) on the status line under the box stops every agent working in the conversation.
- Click the status line to open Working here and stop one agent.
- Stop also appears on a reply while it is being written, and in the Activity panel.
A stopped run shows 🚫 and Stopped. Text it had already written is still posted. A pending approval for that run is refused.
#Reading activity
- The status line under the box says who is working: Researcher is reading files…, A and B are working…, · Developer needs you.
- The Activity button in the header opens Active now (each live run, its trigger and its last three tool steps) and Recent runs (the last 50, with duration, turns, cost when measured, and by an agent for hand-offs). It never shows a model's private reasoning.
- In channels, a 💬 pill on a message counts the replies it drew. Click it to jump to the first.
#Other tools in a conversation
- Find in conversation: Ctrl+F (⌘+F on macOS).
- Reactions: hover a message and choose Add reaction. Your reactions never reach agents.
- Copy message: on hover.
- Tasks: in the Details panel, New task creates a task with a title, description and assignees. Agents read tasks with
get_channel_contextand move their own withupdate_task, which posts Marked title as status. - Spend: the Details panel shows measured spend for this conversation, the last 24 hours and all time, and your Claude Code or Codex plan usage.
#Inbox
Inbox in the sidebar collects agent replies and errors from every channel and DM, newest first. Unread rows are bold. Click one to open the conversation. Flow runs are not included.
There is no message editing, deletion of single messages, or threads in this version.