Agents
Agent Profiles
Save how an agent CLI launches, then start it from a tab menu, New Workspace, the phone, the CLI, or an orchestration run.
On This Page
An agent profile is a named launch recipe for one of Alera’s built-in agent adapters: which CLI starts, with which flags, and what instructions travel with every prompt. The runtime host keeps the catalog, so the desktop, the paired phone, and the alera CLI share it.
Create A Profile#
- Open Settings → Agent Profiles (
Mod+Commaopens Settings;Modis ⌘ on macOS and Ctrl on Windows and Linux) and click New Profile. - Enter a unique Name and pick the Adapter Type, the agent CLI it launches.
- Choose a Launch Mode and set its options.
- Turn on Show In New Tab Menu (off by default) to list it in tab menus.
- Optionally write a Custom Prompt, added after every starting prompt, then click Save.
Under Routing, Description tells an orchestration coordinator what the profile is for. Profiles sharing a Quota Group drain one usage bucket, so a fallback avoids that group.
Managed And Command Profiles#
Managed profiles are built from options, and Command Preview shows the line Alera assembles; the runtime host quotes it for the shell the tab runs. Agent Default leaves an option to the CLI’s own configuration.
Command profiles take a line you write, for flags Alera does not model. Write only the flags; the Prompt Delivery note shows where the prompt goes. Keep it interactive: print modes such as -p, --print, or -x answer once and exit, leaving no agent in the tab.
Test Command runs either kind in a terminal dialog without a task.
All adapters but Amp and fx have a Model option, listed from the installed CLI when possible (Refresh Models reloads; Exact Model ID takes any other). Other options vary:
- Codex: Reasoning Effort, Plan Mode Reasoning Effort, Sandbox, Approval Policy, Web Search, Bypass All Protections.
- Claude Code: CCS Profile, Persona, Reasoning Effort, Permission Mode, Allow Skip Permissions.
- GitHub Copilot: Persona, Reasoning Effort, Mode, Context, Allow All, Maximum AI Credits, Maximum Autopilot Continues, Do Not Ask User.
- Cursor: Effort, Thinking, and Fast model variants, Mode, Review Mode, Sandbox, Trust Workspace.
- Antigravity: Persona, Reasoning Effort, Mode, Skip Permissions, Sandbox.
- OpenCode and OpenCode 2: Persona, Auto Approve. OpenCode 2 launches with
--standaloneand passes only Auto Approve to its interactive TUI. - Pi: Thinking, Project Trust.
- Amp: Mode, Fast Mode.
- Grok Build: Persona, Reasoning Effort, Permission Mode, Sandbox, Disable Web Search.
- fx: Resume Latest Session, Ignore Additional Directories, Record Session.
A CCS Profile runs Claude Code through ccs. Options that loosen protections, such as Bypass All Protections, show a warning, and saving one opens Confirm Reduced Protections.
Default, Clone, And Order#
The star, Set As Default, marks the profile From Prompt preselects on the desktop and the phone; without one, the first profile is used. Clone Profile copies “Codex Sol” as “Codex Sol Copy”, and dragging rows reorders the catalog. Remove checks references first: tabs launched from the profile, automations, and active execution policies block deletion. Each saved profile also has Automation Permissions for automations.
Launching A Profile#
- Tab menu. The Add Tab menu (the + beside the tabs) lists profiles with Show In New Tab Menu on, after Terminal. Pick one, write an Initial Prompt or use Add Files, then Start Agent or Start Without Prompt.
- New Workspace. Press
Mod+Shift+N, choose From Prompt, pick an Agent Profile, then Create And Start Agent. - The phone. The mobile companion offers the same tab menu and From Prompt.
- CLI.
alera agent-profile launch --profile "Codex Sol" --prompt "Fix the flaky test"opens a tab in--workspace, which defaults toALERA_WORKSPACE_ID. - Orchestration. A coordinator dispatches to declared profiles, for example with
alera orchestration delegate --profile "Codex Sol" --spec "Review the tests".
The Starting Prompt#
An agent starts with your prompt, then the profile’s Custom Prompt, then the project’s prompt_append from alera.toml, separated by blank lines.
Alera passes the prompt in the shape each CLI accepts, at launch for every agent but fx:
- Codex, Claude Code, Cursor, and Grok Build: one argument after
--, so a leading dash is not read as an option. - Pi: one bare argument, with a space before a leading dash, because Pi rejects
--. - GitHub Copilot: one
--interactive=argument. - Antigravity: one
--prompt-interactive=argument. - OpenCode and OpenCode 2: one
--prompt=argument. - Amp: standard input, through a generated script.
- fx: pasted once fx reports that it is ready, which needs fx Status (macOS and Linux).
The launch line is typed into your interactive shell, so that shell’s line limit bounds a very long prompt: 8191 characters on cmd.exe.
Resuming A Conversation#
Resume Session in the launch dialog takes a Session ID and opens a new tab on that conversation with the profile’s settings. It sends no prompt; the conversation must exist for that agent and account on the machine running the tab. Alera also resumes by itself after a terminal restarts (CLI Agents), which relies on the agent’s status hook.
From The CLI#
alera agent-profile lists, shows, creates, updates, reorders, removes, and launches profiles. --json goes before the subcommand, and settings that newly reduce protections need --confirm-reduced-protections. Orchestration has examples. The Agent Profiles Skill (Settings → Agents → Extra Skills) teaches an agent to manage profiles.