Memory and Experience

Every agent in matic maintains two distinct knowledge layers: Memory — a mutable, curated store of context carried into each execution — and Experience — an append-only personal work log written at the close of every engagement. Together they allow agents to improve over time, surface relevant past decisions during planning, and feed the Learning phase that drives Archetype capture. This section covers how both layers are structured, how they fit within the execution context envelope, and how matic keeps them accurate.

Context Budget

How matic allocates the execution context envelope across persona, skills, memory entries, active policies, and runtime state — including the truncation priority order that applies when total size exceeds the model's context window limit.

Context Rot Prevention

How stale, contradictory, or irrelevant memory accumulation is detected and remediated to prevent degraded reasoning quality in long-running agents.

Experience Entries

The schema, filesystem location, and lifecycle of Experience entries — the append-only work log each agent writes during the Learning phase, recording what was done, what failed, what was learned, and what the agent would do differently.

Memory Curation

How agents actively maintain their Memory layer between and during engagements: writing new entries, deprecating stale ones, and compressing accumulated context to stay within budget without losing signal.

Memory Scoping and Inheritance

How Memory is scoped to individual agents while providing read access to org-, project-, and team-level Library assets — and what that visibility boundary means for multi-agent teams.

Library

The immutable, PR-gated reference knowledge layer shared across agents at org, project, and team scopes. Covers the Library schema, entry types, filesystem layout, retrieval protocol, indexing, and the promotion and demotion paths that move knowledge between an agent's private Memory and shared Library.