Memory

Clawpy memory is designed to make agent behavior continuous across sessions while keeping operator control and auditability intact.

It combines short-lived execution context with durable retrieval layers so agents can remember prior work without becoming noisy or unbounded.

Two Memory Domains

Clawpy separates memory concerns into two domains:

Personal Continuity

Personal continuity covers user-specific context (preferences, recurring patterns, and prior interactions) that improves assistant behavior over time.

Why this matters: recurring workflows should not require retraining the assistant on every session.

Workspace Execution Memory

Workspace memory tracks project and task context used by execution agents:

Current-step context for active work
Structured operational history
Semantic retrieval over prior work
Durable canonical knowledge for stable project facts

Why this matters: execution quality depends on recall of prior decisions, validations, and constraints.

Retention and Recall

Memory is not treated as one flat store. Clawpy applies layered retrieval and prioritization so the most relevant context is injected first, while less relevant memory remains recoverable.

During rebuild, exact retention values and scoring internals may be tuned, but the operator-facing behavior remains:

Recent and relevant context is prioritized
Durable knowledge stays available across runs
Operator corrections feed future recall quality

Operator Controls

Operators can manage memory behavior through product controls and policy settings: