AlgoFoxGet AlgoFox App

CoursesAgent Harnesses in Practice

Memory Designs

The three ways a harness remembers: a small file loaded every session, a store searched from outside the window, and the flush that saves a fact before compaction rewrites the chat. What to keep, what to drop, and which scope a fact belongs to.

Lessons
6
Exercises
34
Minutes
37

Start The File in the Window

  1. 1

    The File in the Window

    After this lesson you can say what a memory file loaded at session start gives the model, when an edit to it becomes visible, and what happens when it outgrows its budget.

    Fill blankMultiple choiceTap
    5 exercises
    5 min
  2. 2

    Search Outside the Window

    After this lesson you can say which memory stays outside the window, how a search pulls one piece in per turn, and why an outside provider adds to the file rather than replacing it.

    Fill blankMultiple choiceTrace
    5 exercises
    6 min
  3. 3

    Compaction and the Flush

    After this lesson you can say which memory survives compaction, why a fact said once in chat may not, and how three harnesses get a fact out of the conversation in time.

    DebugCode orderMultiple choice
    5 exercises
    6 min
  4. 4

    What to Store

    After this lesson you can sort candidate memory entries into keep and drop, name the scope a fact belongs to, and say why raw dumps and session-only facts do harm.

    Fill blankMultiple choiceTap
    5 exercises
    5 min
  5. 5

    Checkpoint: Memory DesignsCheckpoint

    File, store, flush and scope, on an agent and a repo you have not seen.

    DebugCode orderMultiple choiceTapTrace
    6 exercises
    6 min
  6. 6

    Boss: Remember a CustomerBoss

    Design memory for an agent that must remember a customer across months: what session one learns, file or store, the budget hit, the compaction that would lose a fact, the flush, the search next month, the stale fact, and the handoff.

    DebugFill blankMultiple choiceTrace
    8 exercises
    9 min