multisage（multisage 搜索）

Multisage Skill

Query the Multisage API for multi-expert AI answers. Multisage sends your question to Claude, GPT, and Gemini simultaneously, then synthesizes their responses into a single, well-rounded answer — highlighting where they agree, disagree, and what unique insights each provides.

Prerequisites

The multisage CLI must be installed globally:

npm install -g multisage

An API key is required. Get one at https://multisage.ai/settings (under "API Keys").

API Key Setup (Do This First)

Before running any multisage command, you must ensure MULTISAGE_API_KEY is available in your shell. The env var is often NOT inherited automatically — check and load it explicitly:

# 1. Check if it's already set
echo "${MULTISAGE_API_KEY:0:8}"

# 2. If empty, look for it in common locations
grep MULTISAGE_API_KEY .env 2>/dev/null || grep MULTISAGE_API_KEY ~/.env 2>/dev/null

# 3. Export it (replace with actual key found above)
export MULTISAGE_API_KEY="msk_..."

If you get a timeout or authentication error, the most likely cause is a missing API key — always verify it's set before debugging further.

Running from Claude (Important)

When running multisage from Claude or any non-interactive context, always use the -q flag to suppress the interactive spinner (doesn't render in tool output).

Redirect output to a file with > instead of piping through tee, which can cause issues in non-TTY contexts:

# Correct — redirect to file
multisage -q "your question" > /tmp/multisage-output.txt 2>&1
cat /tmp/multisage-output.txt

# Wrong — tee can cause issues in non-TTY contexts
multisage -q "your question" | tee output.txt  # DON'T DO THIS

Standard Query

For most questions, run a standard query. This consults all three AI providers and returns a synthesized answer with stage breakdown:

multisage -q "your question here"

The default output shows each stage: quick answer, expert synthesis, and debate (if the experts disagree). This costs 1 credit.

Deep Research

For questions that need thorough, in-depth analysis — research topics, complex technical questions, literature reviews — use deep research mode. This launches 3 parallel deep research sessions (one per provider) and synthesizes the results:

multisage -q --deep-research "your question here"

Deep research costs 5 credits and takes 5-25 minutes. The CLI streams progress updates as each provider completes. You can detach with Ctrl+C and check results later.

Output Options

Combine flags as needed. For example, -f -q gives full output without a spinner. -j gives structured JSON (useful if you need to parse the response programmatically).

Managing Deep Research

Deep research runs asynchronously. If you start one and need to check on it later:

# Check if results are ready
multisage results <threadId>

# Cancel an in-progress query
multisage cancel <threadId>

The threadId is printed when you start a deep research query.

When to Use Which Mode

• Standard (multisage -q "...") — Good for most questions. Quick (30-90 seconds), costs 1 credit. Best for factual questions, advice, comparisons, code help.
• Deep Research (multisage -q --deep-research "...") — For thorough analysis. Each provider does its own deep research with web search, then results are cross-referenced and synthesized. Best for research topics, technical deep-dives, market analysis, literature reviews.

Typical Usage from Claude

# Standard query — redirect to file, then read
multisage -q "What are the tradeoffs between SQLite and PostgreSQL for a hobby project?" > /tmp/ms-output.txt 2>&1
cat /tmp/ms-output.txt

# Full output (includes individual expert responses)
multisage -q -f "What are the tradeoffs between SQLite and PostgreSQL for a hobby project?" > /tmp/ms-output.txt 2>&1
cat /tmp/ms-output.txt

# Deep research
multisage -q --deep-research "What are the latest advances in protein folding prediction?"

# JSON output for parsing
multisage -q -j "Is Rust or Go better for CLI tools?" > /tmp/ms-output.txt 2>&1
cat /tmp/ms-output.txt | jq '.answer'

Response Structure

The API returns:

• answer — Final synthesized answer
• experts — List of expert names (or full details with -f)
• creditsUsed — Credits consumed (1 for standard, 5 for deep research)
• stages.quickAnswer — Initial quick answer before expert consultation
• stages.synthesis — Synthesized answer incorporating all expert perspectives
• stages.debate — Debate section highlighting disagreements (null if experts agree)

Limits and Errors

• Questions are truncated to 1000 characters
• 5-minute timeout for standard queries
• API keys must start with msk_
• Rate limit: 10 requests/minute, 3 concurrent

Troubleshooting

Timeout errors are almost always caused by a missing API key — the CLI makes an unauthenticated request that hangs instead of returning a clean error. Fix: verify MULTISAGE_API_KEY is exported (see API Key Setup above).

Diagnostic checklist (run these before retrying):

# Is the API key set?
[ -n "$MULTISAGE_API_KEY" ] && echo "Key set: ${MULTISAGE_API_KEY:0:8}..." || echo "KEY NOT SET"

# Is the CLI installed?
which multisage && multisage -V

# Quick connectivity test
multisage -q "hello" > /tmp/ms-test.txt 2>&1 && echo "OK" || echo "FAILED — check output:" && cat /tmp/ms-test.txt