Commands Reference

Coqui provides two interfaces: an interactive REPL with slash commands, and a CLI with subcommands. This document covers both.

REPL Commands

All REPL commands start with /. Type /help during a session to see a quick reference.

Session Management

Agent & Roles

Group sessions stay orchestrator-managed and clear any single active profile scope. Starting or reshaping a group session reuses the same membership validation and duplicate-composition rules as the HTTP API.

When the active session is group-enabled, the REPL prompt shows the current member list, for example You (group session with @alex-hormozi, @trinity). Prompts without explicit member mentions fan out to all members in stored order. Use @name to narrow the responder set or @everyone / @group to force a full-team reply. During a turn, streamed start, iteration, and reasoning output is labeled with the active member so you can see who is speaking and why that responder set was chosen.

Configuration & System

Tool & Toolkit Management

Backstory Generator

Images

/image generate accepts option flags such as --model=vendor/model, --vendor=openai|ollama, --tags=a,b, --category=name, --file-hint=name, --save-dir=images/custom, --size=1024x1024, and --quality=standard|hd.

Toolkit commands now use the same shared help formatting conventions as the built-in REPL commands. When a toolkit implements structured help metadata, /command and /command help render that metadata through Coqui’s standard formatter; otherwise Coqui generates the page from the command registration data.

When the resolved model is an Ollama image model that is not installed locally, Coqui now asks for confirmation before running ollama pull and shows a controlled download status instead of dumping the raw pull stream into the REPL.

In the interactive REPL, Coqui now auto-renders one low-fidelity ANSI image preview per turn when a successful tool result returns a workspace-local image path, such as a browser_capture screenshot or an image-toolkit output. Streamed assistant markdown also previews one workspace-local markdown image per response. This v1 behavior is intentionally narrow: local workspace files only, first preview only, and graceful fallback when ext-gd is unavailable.

Background Tasks

Planning & Todos

Projects & Sprints

Scheduling & Automation

Inspection commands in this section are user-facing. Mutation and lifecycle controls are still available, but they are best treated as advanced automation or operator workflows rather than routine chat interaction.

Context Management

Options can be combined: /summarize recent 5 focus "database schema".

Marketplace

CLI Commands

Coqui is invoked via coqui. By default, coqui starts the launcher-managed app: REPL in the foreground plus the API in the background. coqui-launcher is the explicit equivalent. Advanced console commands such as doctor, benchmark, task:run, and turn:run remain available when explicitly requested.

coqui

Start the full launcher-managed app or use launcher service controls.

coqui                      # Start REPL + API (default)
coqui --api-only           # Start only the launcher-managed API
coqui --repl-only          # Start only the REPL via the launcher
coqui --wizard             # Run the setup wizard
coqui status               # Show launcher-managed service status
coqui stop                 # Stop launcher-managed services
coqui cleanup              # Clean stale/conflicting Coqui processes

coqui run

Compatibility alias for REPL-only mode via the launcher.

coqui run                      # Start the REPL only
coqui run --new                # Start a fresh session
coqui run --session abc123     # Resume a specific session
coqui run --profile caelum     # Resume or create the last active session for a profile
coqui run --auto-approve       # Skip tool confirmation prompts
coqui run --continue           # Resume last session and auto-send "Continue."

Environment variable equivalents: COQUI_UNSAFE=true for --unsafe, COQUI_AUTO_APPROVE=true for --auto-approve.

Headless Mode

Run a single prompt without the interactive REPL:

coqui run --no-terminal --prompt "List all PHP files in src/"
coqui run --no-terminal --prompt "Summarize this project" --format json
printf '%s' 'Summarize this project with a very large prompt body' | coqui run --no-terminal --format json

For very large prompts, prefer piping via stdin instead of --prompt. Coqui does not truncate headless input, but shell and OS argv limits can prevent very large --prompt values from reaching the process intact.

coqui api

Compatibility alias for launcher-managed API-only mode.

coqui api                             # Start on 127.0.0.1:3300
coqui api --host 0.0.0.0 --port 8080  # Listen on all interfaces, port 8080

Environment variable: COQUI_API_HOST overrides the --host default when the flag is not explicitly passed.

coqui setup

Compatibility alias for the launcher setup wizard.

coqui setup                       # Configure in current directory
coqui setup --workdir /my/project  # Configure a specific project

coqui doctor

Run health checks on your Coqui installation.

coqui doctor              # Run all checks
coqui doctor --repair     # Fix detected issues automatically
coqui doctor --json       # Output results as JSON

coqui benchmark

Run lightweight performance benchmarks against key Coqui subsystems such as boot, token estimation, SQLite setup, and autoloader hot paths.

coqui benchmark
coqui benchmark --iterations 250
coqui benchmark --json

Launcher

coqui-launcher is the explicit launcher name for the same public entrypoint surface exposed through coqui.

Modes

coqui-launcher                   # Start REPL (foreground) + API (background)
coqui-launcher --repl-only       # Start REPL only, no API
coqui-launcher --api-only        # Start API only (foreground)
coqui-launcher --wizard          # Run setup wizard directly
coqui-launcher stop              # Stop all background services
coqui-launcher stop-api          # Stop the background API only
coqui-launcher status            # Show which services are running

Launcher Flags

Exit Code Behavior

Signal Handling

At the REPL prompt

The counter resets after any successful input.

During agent execution

See Also

• ROLES.md — Built-in roles and how to create custom roles
• FEATURES.md — Complete feature reference
• CONFIGURATION.md — Configuration file reference
• API.md — HTTP API endpoints