Skip to main content
Coding agents are the AI coding tools you already use, such as Claude Code and Codex, run inside Enconvo chats. Use one when the work is code in a project folder: it edits files with diffs, runs commands, and asks before it acts, while Enconvo keeps the chat, its history, and its approvals in the Chat Window, SmartBar, and your phone.

Before you start

Each coding agent has its own page in Settings → Agent List. Click it to see its Status:
  1. If it shows Not installed, click Install, or run the command shown in Terminal and click Check again. A version that’s too old shows Update.
  2. Under Account, click Sign in with Claude, Sign in with ChatGPT, Sign in with Grok, Sign in with Google, or Sign in with Meta. Enconvo uses the same sign-in as the agent’s own command line in Terminal.
When the Claude Code or Codex sign-in ends within ten days, the agent’s page, the Updates list, and a Project’s Board say so, with Sign in again. A Board task whose Coding agent is signed out waits as Held and runs again once you sign in. Muse Code running on its local model needs no Meta sign-in. Grok Bot needs a routine instead; see Connect Grok Bot.

Chat with a coding agent

  1. In the Chat Window, pick the coding agent from the Agent cards or the Agent menu.
  2. Choose the project folder it works in.
  3. Type your request and send. Its steps, diffs, and commands stream into the chat, and its approval questions show as cards for you to answer.
The chat’s approvals follow Enconvo’s Permission mode. At Full access, a new Claude Code chat starts in Bypass permissions (where Claude Code’s own settings allow it) and a new Codex chat in Full access, so neither asks before it acts. In a worktree chat, the side panel’s Pull request tab (⌃⇧P) shows the branch’s pull request on GitHub, or creates one. See Chat Window. To pick up an earlier session, choose Resume a previous session. Sessions you started in the agent’s own command line are listed too, under the names you gave them. You can also hand a chat you started with Mavis to a coding agent by continuing in a new chat with it. Other Agents can use coding agents as subagents, and a Project can give them tasks or have one lead the Project. See Subagents and Projects.

Claude Code: effort, fast mode, and Ultracode

The effort picker in a Claude Code chat sets its reasoning effort. When the session supports it, the picker also shows Fast mode and an Ultracode switch. Ultracode turns on workflow orchestration for the rest of the session at whatever effort you pick, and stays on when you change the effort. Typing /effort ultracode does the same. In a new chat, you can set fast mode and Ultracode before the first message.

Codex: warnings

Codex’s warnings stay out of the conversation. Type /warnings to list them.

Muse Code: continue in Terminal

To continue a Muse Code chat in Terminal, choose Continue in Terminal, or Copy command to copy it. The chat opens with muse resume in its folder. When you quit Muse Code there and send from Enconvo again, the turns you ran in Terminal come back into the chat. Muse Code’s approval questions show the change an edit would make. When you deny a step, what you type with Deny goes to Muse Code in the same turn, so it can act on it right away. Its skills and /compact are in the / menu.
Muse Code’s Contributor mode is on by default, which lets Meta train its models on your prompts and code. Enconvo leaves it as it is; turn it off in Muse Code’s own settings.

Choose what a coding agent can use in Enconvo

Coding agents can use Enconvo’s plugins in their chats, such as messaging your Enconvo Agents, image and video creation, memory, the browser, and Computer Use. On the agent’s page in Settings → Agent List, under Enconvo Capabilities:
  • Turn off Enconvo Capabilities to keep that agent away from all of Enconvo’s plugins.
  • Or click Customize next to Plugins and turn off one plugin at a time. A plugin that’s off is hidden from that agent, and its calls are refused.
The change applies to the agent’s open chats too. Coding agent chats in a client’s Workspace stay entirely inside that Workspace.

Connect Grok Bot

Grok Bot runs in xAI’s cloud. Enconvo sends each message to a Webhook routine you make in Grok Bot, and shows the reply in the chat, even when it arrives late.
  1. In Grok Bot, create a routine with a Webhook trigger.
  2. Open Grok Bot in Settings → Agent List, click Copy next to Routine instructions, and paste them into the routine.
  3. Add the routine’s webhook URL and key as the credential on the same page, then click Test. The test waits for the routine’s reply for up to about 90 seconds.
Until the routine is connected, the agent picker shows Grok Bot as Not set up. Every message uses your Grok Bot plan’s weekly usage. Stop only stops Enconvo waiting: the routine’s run goes on in the cloud, and a reply that comes later is still added to the chat. A message can be up to 32,000 characters. Attached files and images stay on your Mac; only the text is sent.

Let Grok Bot call Enconvo

With Enconvo’s MCP server, Grok Bot can reply to your chats, notify your Mac and your paired phones, and ask an Agent on your Mac to run a task.
  1. On Grok Bot’s page, under Grok Bot calling Enconvo, click Connect. You need to be signed in to your Enconvo account.
  2. In Grok Bot, open Plugins, add the Server URL shown, and sign in with your Enconvo account when Grok Bot asks.
Each request to an Agent on your Mac waits for your approval, unless you turn on Always allow ask_mac. Requests stay listed in the Grok Bot on this Mac chat. Click Disconnect to stop Grok Bot from calling Enconvo; disconnecting, or your Mac going offline, withdraws a request still waiting.

Troubleshooting

Agents

Subagents, roles, and approvals.

Projects

Give coding tasks to a team in one folder.

Chat Window

Start, steer, and organize chats.

Hooks

Run your own commands when an Agent acts.