@llm-flow/multi-model

Run a prompt against multiple AI provider CLIs in parallel and persist the results as a session.

Ships three surfaces backed by the same library core:

• multi-model query — primary CLI subcommand
• multi-model mcp — stdio MCP server exposing query, doctor, list-providers
• multi-model doctor — checks provider CLIs are on PATH

Install

npm install -g @llm-flow/multi-model

Use

# Run a prompt against three providers in parallel
multi-model query \
  --providers claude,gemini,codex \
  --prompt-file ./review-prompt.md \
  --session-root .llm-flow/sessions

# Confirm providers are installed and authenticated
multi-model doctor

# Start the MCP server (stdio)
multi-model mcp

Output

Each query writes to <session-root>/<session-id>/:

prompt.txt           verbatim prompt sent to all providers
claude.md, claude.json
gemini.md, gemini.json
codex.md,  codex.json
metadata.json        session metadata (start/end times, durations, errors)

Library

import { fanOut, doctor } from '@llm-flow/multi-model';

const result = await fanOut({
  providers: ['claude', 'gemini'],
  prompt: 'Review this diff for security issues...',
  sessionRoot: '.llm-flow/sessions',
  profile: 'max-quality',
});

console.log(result.sessionDir);
console.log(result.results.map((r) => `${r.provider}: ${r.success ? 'ok' : r.error}`));

Provider CLIs assumed

| Provider | CLI | Max-quality model flag | |---|---|---| | claude | claude --print --output-format text | --model opus | | gemini | gemini | -m gemini-3.1-pro-preview | | codex | codex exec | -m gpt-5.4-codex |

If a provider CLI isn't on PATH, query reports it as failed and continues with the others. Run multi-model doctor to see what's missing.

License

MIT