neuralmind

Use Case: Any LLM (ChatGPT / Gemini / Local Models)

What you’re solving for

You want NeuralMind-quality context but you’re not using Claude Code, Cursor, or any MCP-compatible agent. You’re in ChatGPT, Gemini, a Perplexity thread, an Ollama session, or just a custom script. You need portable context that fits any chat window.

The pattern

NeuralMind’s CLI output is plain text. Pipe it into any clipboard, file, or request body — the model doesn’t care how it got there.

Setup

pip install neuralmind
cd your-project
neuralmind build .

Patterns

1. Copy-paste into a web chat

macOS:

neuralmind wakeup . | pbcopy
# paste into ChatGPT / Gemini / Claude.ai

Linux:

neuralmind wakeup . | xclip -selection clipboard

Windows (PowerShell):

neuralmind wakeup . | Set-Clipboard

Then start your chat with: “Here is the project context: <paste>. My question is …”

2. Piped into a one-shot script

CONTEXT=$(neuralmind query . "How does auth work?")
curl https://api.openai.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d "$(jq -n --arg c "$CONTEXT" --arg q "explain the flow" '{
    model: "gpt-4o",
    messages: [
      {role: "system", content: $c},
      {role: "user", content: $q}
    ]
  }')"

3. Saved as a file the model reads

neuralmind wakeup . > CONTEXT.md
neuralmind query . "How does billing work?" > BILLING_CONTEXT.md

Attach those files to a Gemini Project, Claude Project, or Ollama system prompt. You get NeuralMind retrieval without needing the MCP server running.

4. Per-question context

Don’t load the whole repo — load just what the question needs:

neuralmind query . "How does the payment webhook validate signatures?" | pbcopy

Paste, ask, done. Usually ~800–1,100 tokens instead of tens of thousands.

What you lose vs. Claude Code

• No PostToolUse compression — that’s Claude-Code-specific. You still get the retrieval-side savings (40–70×), just not the additional Read/Bash/Grep compression layer.
• No MCP tool calls — the model can’t invoke neuralmind_query itself; you invoke it and paste.

What you still get

• The same 4-layer progressive disclosure
• The same skeleton views (neuralmind skeleton <file>)
• The same semantic search (neuralmind search . "term")
• Fully local, no data leaves your machine
• Works with any model — OpenAI, Google, Anthropic, Ollama, vLLM, anything

Graph view works with any LLM (v0.6.0+)

The graph view (neuralmind serve) is agent-agnostic — it doesn’t care which chat window you’re piping context into. If you use ChatGPT, Gemini, or a local model:

1. neuralmind serve . in a terminal.
2. Browser shows the live graph.
3. Run neuralmind query . "..." in a second terminal — its nodes pulse on the canvas as the query executes, even though ChatGPT itself never called NeuralMind directly.
4. Pipe the CLI output into your chat window.

The pulse-rings feed reflects every CLI call, every neuralmind watch daemon event, and every file edit — regardless of which LLM you ultimately send the context to. The trust-gap closure benefit (see claude-code.md) applies anywhere you’re using NeuralMind retrieval.

Pro tip: system prompt pattern

Generate context once at the start of a long session:

neuralmind wakeup . > /tmp/sys.md

Set the contents of /tmp/sys.md as your system prompt. For specific questions during the session, append neuralmind query . "..." output to the user message.

---

← Back to use-case index · Main README

This site is open source. Improve this page.