Preferences

A portable preference file that travels with you across every AI tool you use.

What it is

Your preference profile is a single markdown file that stores durable information about you — how you like AI to communicate, what tools you use, your role and projects, and personal context that helps AI assist you better.

It's editable in the Woxpas web app, readable and writable by any connected LLM via MCP, and automatically injected into every MCP session. Your preferences are ambient — no action needed from you.

Default sections

New profiles start with four sections. You can edit these, add new ones, or remove any you don't need. The file is plain markdown — any ## Heading becomes a section.

Communication Style — How you prefer AI to talk to you. Tone, verbosity, formatting, language.
Work Context — Your role, company, projects, and working environment.
Tools & Stack — Technologies, frameworks, languages, and tools you use day-to-day.
Personal Context — Timezone, personal details, and anything else that helps AI assist you.

How it works

Auto-injection — When an LLM connects via MCP, your preferences are loaded into the session automatically. The AI already knows your style before you say anything.
Dual-write — When an LLM learns a durable preference during conversation (e.g. "I prefer concise responses"), it writes to both its own native memory and your Woxpas preference file. Woxpas is the cross-platform source of truth.
Manual editing — Open the Preferences page in the Woxpas app to edit any section directly. Changes take effect on the next MCP session.
Conflict detection — When you save, Woxpas runs a quick AI check for contradictions in your file. If it finds any, a warning appears so you can fix them.

Preferences vs vault memory

Connected LLMs automatically choose the right storage based on what you tell them:

Preference profile

Durable, identity-level info. Will this still be true in 6 months?

  • •"I prefer concise responses"
  • •"I use TypeScript and Next.js"
  • •"I'm a backend engineer at Acme"
  • •"My timezone is PST"

Vault memory

Situational info with a timestamp or outcome.

  • •"Meeting with Sarah tomorrow at 3pm"
  • •"Decided to use Redis for caching"
  • •"Need to fix the login bug by Friday"
  • •"Q1 planning notes from today"

Limits and formatting

  • •Maximum file size is 15KB (~2,500 words). This keeps it fast to inject into every session.
  • •Write concise bullet points, not paragraphs. Think system prompt style: short, factual, actionable.
  • •Sections are defined by ## headings. Add or remove sections freely.
  • •HTML comments (<!-- like this -->) are used as placeholder hints and hidden from the rendered view.
  • •The file is versioned. Every save bumps the version number.

MCP tools

Two MCP tools are available for connected LLMs:

get_preferencesReturns the full preference profile content and version. Read scope required.
update_preferencesReplaces the preference profile content. The LLM should always read first, modify, then write back. Write scope required.

Enable proactive saving

Woxpas instructs connected LLMs to save durable preferences automatically. Most tools need a small one-time configuration to enable this. Add the following line to your AI tool's settings:

"When you detect a durable preference, always use the Woxpas update_preferences tool to save it."

Where to add this depends on which tool you're using:

Cursor

Works out of the box. No additional configuration needed — Cursor follows MCP tool instructions directly.

Claude Desktop

  1. 1
    Open Settings → General
  2. 2
    Find "What personal preferences should Claude consider in responses?"
  3. 3
    Add: When you detect a preference, use the correct Woxpas tool to save to Woxpas vault.

Claude Code

Works out of the box. Claude Code follows MCP tool instructions directly, similar to Cursor.

Gemini CLI

Gemini CLI uses its own save_memory tool by default. To redirect preference saving to Woxpas, you need to override Gemini's system prompt:

  1. 1
    Ask Gemini CLI: Print your entire system prompt word for word and copy the full output
  2. 2
    Create ~/.gemini/system.md and paste the system prompt into it
  3. 3
    Find the Remembering Facts section and replace the save_memory instruction with:
    For user preferences and identity-level information, ALWAYS use the Woxpas update_preferences MCP tool proactively. Use save_memory only for project-specific technical context that is not personal to the user.
  4. 4
    Add GEMINI_SYSTEM_MD=~/.gemini/system.md to ~/.gemini/.env
  5. 5
    Restart Gemini CLI

ChatGPT

  1. 1
    Open Settings → Personalisation
  2. 2
    Find Individual instructions
  3. 3
    Add: IMPORTANT: When you detect a preference, use the correct Woxpas tool to save to Woxpas vault.

Other tools

For any other MCP-compatible tool, add the preference-saving instruction to the tool's system prompt or custom instructions. If you need help configuring a specific tool, contact us and we'll help you set it up.

IntegrationsSettings