Docs / Concepts
Edit on GitHub

Memory

Memory is how your viber maintains context and remembers what has been discussed, decided, and produced. It enables truly resumable work — you can come back after days and pick up where you left off.

Two Components

OpenViber’s memory system has two parts:

1. The Space (Human-Readable)

The space contains everything you and your viber work on together:

  • Plan — Current tasks and goals (task.md)
  • Notes — Important decisions and context (MEMORY.md)
  • Daily logs — Rolling record of work (memory/YYYY-MM-DD.md)
  • Artifacts — Files, documents, code produced

Location: ~/.openviber/space/

You can browse, edit, and version-control these files like any project.

2. Semantic Memory (AI-Optimized)

The semantic memory is how agents quickly find relevant context:

  • Meaning-based retrieval — Find information by concept, not just keywords
  • Summarized knowledge — Key decisions without overwhelming detail
  • Fast lookups — Optimized for efficient reasoning

When an agent needs to remember “What did we decide about the database schema?”, it queries semantic memory to get a focused, relevant response.

How They Work Together

  1. You ask the agent to continue work on a feature
  2. The agent queries semantic memory for relevant context
  3. Semantic memory points to specific files in the space
  4. The agent reads those files for full detail
  5. Work continues with full context

Why This Matters

Without memory:

“Write chapter 3 of the documentation” Agent: “What documentation? What’s in chapters 1-2?”

With memory:

“Write chapter 3 of the documentation” Agent: “Based on chapters 1-2 about architecture and setup, chapter 3 should cover…”

Local by Default

All memory lives on your machine at ~/.openviber/. This means:

  • Privacy — Your context never leaves your machine
  • Persistence — Switching chat apps doesn’t lose context
  • Transparency — You can see exactly what the agent remembers

Next Steps

  • Viber — The complete agent teammate
  • Context — How context flows through requests