multi-ai-consult

# Multi-AI Consult

Consult multiple AI assistants in parallel and get a synthesized response.

## Features

- **Parallel Execution**: Query Gemini, Codex, Qwen, and OpenCode simultaneously
- **Platform Detection**: Automatically uses `gtimeout` on macOS, `timeout` on Linux
- **Full PATH Resolution**: Resolves CLI binary paths at startup to prevent exit 127 errors
- **Response Truncation**: Prevents 25K token overflow with intelligent truncation
- **Tiered Synthesis**: Adapts strategy based on response size (Small/Medium/Large)
- **Enhanced Error Classification**: Categorizes errors (auth, rate limit, network, timeout) for smart retry
- **Graceful Degradation**: Works with any subset of available AI CLIs

## Prerequisites

Install at least one AI CLI:
- `gemini` - Google Gemini CLI
- `codex` - OpenAI Codex CLI
- `qwen` - Alibaba Qwen CLI
- `opencode` - OpenCode CLI

Also required:
- `jq` - JSON parsing
- `coreutils` (macOS): `brew install coreutils` for `gtimeout`

## Usage

```bash
# Basic consultation
/consult <your question>

# Query specific AIs only
/consult --only=gemini,codex <question>

# Exclude specific AIs
/consult --exclude=qwen <question>

# Show individual responses before synthesis
/consult --verbose <question>

# Preview prompt without executing
/consult --dry-run <question>
```

## Configuration

| Variable | Default | Description |
|----------|---------|-------------|
| `CONSULT_TIMEOUT` | `120` | Timeout per CLI (seconds) |
| `CONSULT_MAX_RESPONSE_CHARS` | `20000` | Max chars per response (~5K tokens) |
| `CONSULT_GEMINI_MODEL` | `gemini-3-pro-preview` | Gemini model |
| `CONSULT_CODEX_MODEL` | `gpt-5.1-codex-max` | Codex model |
| `CONSULT_OPENCODE_MODEL` | `anthropic/claude-opus-4-5` | OpenCode model |

## How It Works

1. **Verify**: Detects platform and resolves full paths to CLI binaries
2. **Prepare**: Constructs a clear prompt from your query
3. **Execute**: Runs all available CLIs in parallel with timeout
4. **Retry**: Failed CLIs get one retry with exponential backoff
5. **Parse**: Extracts text responses using jq (handles NDJSON formats)
6. **Synthesize**: Combines all responses into a unified answer

## Exit Codes

| Code | Meaning |
|------|---------|
| 0 | Success - at least 2 AIs responded |
| 1 | Partial success - only 1 AI responded |
| 2-5 | All AIs failed (auth/rate limit/timeout/other) |
| 10 | No AI CLIs installed |
| 11 | Missing required dependency |

## License

MIT