First Steps

Once installed, here's how to confirm everything works and start using Ask LLM.

Test the Connection

Use ask-llm ping to test the connection

You should see a Pong! reply listing the providers your server detected.

If something's off, run the doctor from your terminal — it works even when the MCP server can't start:

npx ask-llm-mcp doctor

Single-Provider Calls

The unified ask-llm tool takes a provider parameter. In natural language:

Use ask-llm to ask Codex to refactor src/auth.ts
Use ask-llm to ask Antigravity to debate this approach
Use ask-llm to ask Ollama to summarize this file (runs locally, no data sent anywhere)

If you installed the per-provider packages instead of the orchestrator, the tools are named explicitly:

ask codex to review the staged changes
ask ollama to explain this auth flow
ask gemini to summarize @README.md

Multi-Provider Dispatch

Send the same prompt to multiple providers in one call and compare:

Use multi-llm to ask Codex and Gemini whether this approach is thread-safe

Returns per-provider responses + token usage in one structured payload. Per-provider failures are isolated (one provider hitting quota doesn't fail the whole call).

Multi-Turn Conversations

Every response includes a session ID. Pass it back to continue:

Use ask-llm to ask Codex to review src/auth.ts for security issues
# → Response includes [Session ID: abc-123-...]

Use ask-llm to ask Codex to fix the issue you found, sessionId abc-123-...
# → Codex remembers the prior review

Gemini, Codex, and Ollama support sessions — Gemini and Codex use native CLI resume, Ollama uses server-side conversation replay (Antigravity is single-turn). See Multi-Turn Sessions for details.

Interactive REPL

For quick sanity checks without setting up an MCP client:

npx ask-llm-mcp repl

Multi-provider shell with /provider <name>, /new, /usage, /sessions, /help, /quit slash commands. Streams Gemini responses live.

Next Steps

• How to Ask — natural language patterns and the @ file syntax
• Strategies & Examples — proven workflows for code review, debugging, and architecture analysis
• Multi-Turn Sessions — continue conversations across calls
• Plugin Overview — slash commands and subagents for Claude Code users
• Model Selection — Pro vs Flash, GPT-5.5 vs mini, Ollama model choices