Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.4stax.com/llms.txt

Use this file to discover all available pages before exploring further.

4StaX memory is a persistent, ranked context layer that travels with you across every AI provider and session.

​
The problem it solves

Every AI session starts blank. You re explain your stack, your preferences, and your project context every time, to every tool. Multiply that by how many sessions you have per day and it becomes a tax on how effectively you can work with AI. Memory fixes this by maintaining a structured vault of what matters and surfacing the right subset at the right time.

​
How it works

1

Capture

During or after a session, significant context is stored as discrete memory entries, such as preferences, decisions, facts, and project context.
2

Rank

When a new session starts, kontxt ranks your stored memories against the current task using semantic similarity, recency, frequency, and importance.
3

Inject

The top-ranked memories are injected into your prompt automatically via MCP. Your AI client already knows the relevant context before you say anything.
4

Compound

Over time your vault grows. The more sessions you have, the more accurately kontxt can surface what’s relevant. Memory compounds.

​
Cross-provider

The same vault works with any MCP compatible client. Switch from Cursor to Claude Desktop to any future MCP host and your memory follows because it lives in your vault, not in any one provider system.

​
What gets stored

Not everything. Memory is not a transcript archive. What gets stored is structured, classified, and intentional:
  • Preferences. How you like to work, what tools you use, formatting choices.
  • Facts. Your name, role, project names, team context.
  • Projects. Durable project context that should persist across sessions.
  • Decisions. Architectural choices, tradeoffs, and rationale.
  • Skills. What you can do and how you do it.
  • Episodic. Session specific context worth keeping.

​
What should not be stored

Local first does not mean store everything. Some data is better kept out of memory. Avoid storing:
  1. Secrets such as API keys and passwords.
  2. Large logs and raw transcripts.
  3. Highly sensitive personal data unless you explicitly want it in the local vault.

​
How to get good results

Memory works best when entries are short and stable. Guidelines:
  1. Store decisions as single statements with the reason.
  2. Store preferences as plain language.
  3. Use project scoping when the memory is tied to one repo.
  4. Keep episodic entries for things you want later but that are not permanent.

​
Privacy

With kontxt running locally, nothing leaves your machine. The vault is a SQLite file on your disk. You can inspect it, back it up, or delete it at any time.

Get started

Run memory locally in under 5 minutes