Playground

AI memory playground

Nine facts about an invented user called Maya sit in the doko. Ask the agent something and see which ones it would carry into the prompt. Then add your own and try to fool it.

The doko

0 memories · about 0 tokens

    Now ask the agent something

    This demo matches keywords with a small synonym list, in your browser. Production systems match by meaning. Token counts are estimates. Nothing you type is sent or stored.

    What to try

    1. Click Book dinner for Maya on Friday. One memory lights up: the one about food.
    2. Click Plan her trip to Tokyo. Two different subjects are recalled: flights and Japanese.
    3. Ask about something the doko does not hold, such as her favourite film. Nothing is recalled, which is the right result.
    4. Add a memory that answers it, ask again, then press Forget.

    What it leaves out

    A real memory system also has to decide what to write, match by meaning and not only by word, weigh how recent and how important a note is, and resolve contradictions. See how the full loop works or the step-by-step guide.

    About the playground

    What does the memory playground show?

    It shows the recall step of agent memory. You ask a question, and the notes that match are picked out and placed in a short prompt, while the rest are left behind.

    How does the playground choose which memories to recall?

    It scores each memory by the keywords it shares with your question, using a small list of related words, and weights rarer words more. Real memory systems usually compare meaning with embeddings.

    Is a language model used?

    No. The playground is plain JavaScript that runs in your browser. There is no model call and no server.

    Is anything I type saved?

    No. Memories you add stay in the page and disappear when you reload or close it.

    How are tokens counted?

    As a rough estimate of one token per four characters of English text. Real tokenizers differ by model.