CLI Reference

The Consilium CLI is a single npm package (@myconsilium/cli) that runs multi-AI debates from your terminal. It ships full feature parity with Claude Code, Cursor CLI, Gemini CLI, and Grok Build - plan mode, hooks, sub-agents, sandbox, worktree, voice dictation, image generation, MCP marketplace, and headless JSON/stream-JSON output - plus the unique addition of cross-model debate across seven providers.

Run deliberations from your terminal

CLI unit tests: 962
Slash commands: 50+
Deliberation modes: 8
MCP marketplace: 12 seeded
Providers: 7
Dependency CVEs: 0
Quick debate cost: $0.001
Deep debate cost: $0.50

“Plan mode lets the assistant research, read files, and propose an approach without making any changes. You stay in control of when execution begins.”

— Plan-mode framing common to Anthropic Claude Code and Cline docs; Consilium ships an equivalent /plan slash command plus--plan CLI flag.

How do I install the CLI?

curl -fsSL https://install.myconsilium.xyz | sh

Or install directly with your package manager:

npm install -g @myconsilium/clipnpm add -g @myconsilium/cli

What does a quick start look like?

consilium login
consilium config set anthropic-key sk-ant-...
consilium config set openai-key sk-...
consilium debate "How should I architect my microservices?" --mode council

Which commands does the CLI ship?

consilium debate <topic>

Start a multi-agent debate on any topic. Models argue, critique, and synthesize consensus across multiple rounds.

consilium debate "How should I architect my microservices?" --models claude,gpt4o,gemini --mode council --rounds 3

--models <list>	Comma-separated models
--mode <mode>	council, blind, jury, deep, market, redteam, swarm, oracle
--output <fmt>	text, markdown, cursorrules, claude-md, json
--rounds <n>	Number of debate rounds

consilium ask <topic>

Alias for debate. Quick way to get multi-model consensus on a question.

consilium ask "What is the best state management for React in 2026?"

consilium redteam <content>

Run adversarial red-team assessment. Models attack and defend to find vulnerabilities in prompts, plans, or architectures.

consilium redteam "Our authentication flow uses JWT stored in localStorage" --models claude,gpt4o --categories security,injection

--models <list>	Models for attack/defense
--categories <list>	Attack categories to test

consilium eval <topic>

Blind evaluation where model identities are hidden. Responses are anonymized and judged purely on quality.

consilium eval "Explain quantum computing to a 10-year-old" --responses answers.json --models claude,gpt4o,gemini

--responses <file>	JSON file with pre-generated responses
--models <list>	Models to evaluate

consilium benchmark

Run standardized benchmarks across models and deliberation modes. Compare performance on MMLU, TruthfulQA, and HumanEval.

consilium benchmark --benchmark mmlu --models claude,gpt4o --mode council -n 50 --output results.json

--benchmark <name>	mmlu, truthfulqa, humaneval
--models <list>	Models to benchmark
--mode <mode>	Single deliberation mode
--modes <list>	Multiple modes to compare
-n <count>	Number of questions to run
--timeout <ms>	Timeout per question
--output <file>	Save results to file
--local	Use local agent server

consilium chat

Interactive REPL with session persistence, 20+ slash commands, and file/image support. Your full deliberation workbench.

consilium chat --mode council --models claude,gpt4o,gemini

consilium login

Authenticate via browser using Clerk. Token is stored in ~/.consilium/config.json.

consilium login

consilium debug <debateId>

Full diagnostic trace of a debate including timing, token usage, model responses, and error details.

consilium debug dbt_abc123

consilium logs <debateId>

Query structured logs for a specific debate or deliberation.

consilium logs dbt_abc123 -l warn

-l, --level <level>

Filter by log level (debug, info, warn, error)

consilium stats

Model performance dashboard showing usage, costs, response times, and quality metrics.

consilium stats

consilium sessions

Manage chat sessions. List, resume, rename, or delete persistent sessions.

consilium sessions list

consilium config

Set, get, or list configuration values including API keys and defaults.

consilium config set anthropic-key sk-ant-...

Which slash commands work inside chat?

Available inside the interactive chat REPL. Type any command to execute.

Command	Description
/ask <topic>	Start a new deliberation inline
/mode <mode>	Switch deliberation mode
/estimate	Estimate cost of current topic
/output <format>	Change output format
/file <path>	Attach a file for context
/image <path>	Attach an image for analysis
/workspace <path>	Set working directory for file references
/context	Show current context window
/clear	Clear conversation history
/status	Show active deliberation status
/models	List or change active models
/save <name>	Save current conversation
/history	Show conversation history
/conversations	List all saved conversations
/sessions	List chat sessions
/search <query>	Search across all sessions
/rename <name>	Rename current session
/delete <id>	Delete a session
/api	Show raw API request/response
/help	Show all available commands
/exit	Exit the chat REPL

Which output formats are supported?

Control how results are formatted with --output or the /output slash command.

Format	Description
text	Pretty-printed terminal output with colors and formatting
markdown	Full document with metadata, rounds, and synthesis sections
cursorrules	Numbered rules extracted from consensus, ready for .cursorrules
claude-md	Structured Decisions, Guidelines, and Context sections for CLAUDE.md
json	Complete metadata, rounds, messages, and synthesis in JSON

How do I manage chat sessions?

Chat sessions are persisted to ~/.consilium/sessions/*.json and can be resumed, searched, renamed, or deleted.

consilium sessions list
consilium sessions resume sess_abc123
consilium sessions rename sess_abc123 "Architecture Discussion"
consilium sessions delete sess_abc123

Sessions automatically save conversation history, model selections, mode, and attached files. Use /search <query> inside chat to search across all sessions.

Which global flags apply to every command?

Flag	Description
--mode <mode>	Deliberation mode (council, blind, jury, deep, market, redteam, swarm, oracle)
--models <list>	Comma-separated list of models to use
--rounds <n>	Number of debate rounds (default: 3)
--output <format>	Output format: text, markdown, cursorrules, claude-md, json
--verbose	Show detailed round-by-round output
--no-stream	Disable streaming (wait for final result)
--api-url <url>	Custom API server URL
--timeout <ms>	Request timeout in milliseconds
--json	Force JSON output for scripting
--quiet	Suppress non-essential output
--version	Show CLI version
--help	Show help for any command