Consensus before commit.

Every AI model has blind spots. Conclave runs multiple AI CLIs in parallel and surfaces where they agree. When 2 out of 3 models flag the same issue, it's probably real. When only one does, it might be noise.

View on GitHub →
Claude
Codex
Gemini
Qwen
Mistral
Ollama
Grok
How it works
One command. Multiple perspectives. Clear signal.
Terminal — ~/dev/my-project
❯ /review

⊛ conclave — reviewing staged changes

Claude reviewing...
Codex reviewing...
Gemini reviewing...
Qwen reviewing...
Mistral reviewing...
Ollama reviewing...

✓ All reviews complete

2/3 consensus Race condition in fetchData()
Claude and Codex both flagged this. Fix it? (y/n)
Extensible
Build your own commands
~/.claude/commands/my-command.md
# Write your prompt to a temp file
cat > /tmp/prompt.md << 'EOF'
Analyze this and suggest improvements.
$ARGUMENTS
EOF

# Run all configured models in parallel
bash ~/.claude/scripts/conclave-run.sh \
--scope my-command \
--prompt /tmp/prompt.md

# Returns JSON with each model's output
# → {"tools_run": ["codex", "gemini"], "results": {...}}

The engine is a shell script. The commands are markdown files. Configure which models to use, write a prompt, call the script. That's it.