ACT-R memory

Achiral uses several frontier-grade and provisionally patented algorithms on top of an ACT-R-inspired architecture for collective memories. Turn team activity into an emergent memory that can be activated, retrieved, reinforced, decayed, consolidated, and reused over endless sessions.

Remember, the goal of a memory is continuity, whereas the goal of persistence is total recall. Read more about Emergent Memory Systems vs. Persistence Storages.

Memory types

Type	What it stores	Behavior
Procedural	Reusable workflow patterns, tool routines, and action habits	Guides what to do next, subject to capability and review boundaries
Episodic	Recent conversations, events, handoffs, tasks, and work moments	Decays faster unless reinforced
Semantic	Decisions, preferences, domain facts, process knowledge, and stable patterns	Decays more slowly and carries reusable meaning
Flashbulb	High-salience moments that may deserve review	Candidate memory, not automatic truth
Core	Reviewed long-term context, durable documents, and important operating facts	Preserved with stronger governance

These are product memory types, not biological claims. They let Achiral treat a passing chat message differently from a durable policy, an approved decision, or a repeated workflow pattern.

Memory layers

Layer	Role
Signals	Conversations, documents, tasks, connectors, preferences, and workflows enter the system
Traces	Achiral stores scoped memory objects with source, salience, ownership, and permission context
Activation	Memory is ranked by relevance, recency, frequency, salience, tier, and contextual fit
Retrieval	Chiro and personal EAs bring the right memory into the current task
Consolidation	Reused or high-salience memory can move toward semantic or core memory
Governance	Owners and admins review durable memory and preserve organization boundaries

Retrieval lifecycle

Achiral combines search relevance with ACT-R-style activation metadata. A memory can rank highly because it is semantically relevant, because it has strong activation, or because both signals agree.

• Activation: useful memory becomes easier to bring back.
• Decay: stale memory loses weight unless it remains useful.
• Reinforcement: reused memory becomes more available.
• Consolidation: repeated or high-salience memory can become more stable.
• Review: core memory requires human approval before it becomes durable.

Memory vs RAG

Retrieval-augmented generation retrieves context for an answer. Achiral memory adds a lifecycle around that context: formation, retrieval, reinforcement, decay, validation, and auditability.

Dimension	RAG	Achiral memory
Primary job	Retrieve relevant content	Preserve, activate, and reuse operational context
Time horizon	Usually request-time	Compounds across days, people, and workflows
Governance	Depends on the app	Tenant boundaries, roles, validation, and audit logs

Read next

• Memory API for the API surface behind memory operations.
• Procedural memory, episodic memory, and core memory for the main memory-type pages.
• Account settings for the user-facing memory surface.
• Organization for shared organization-level memory and governance.

Learn more

• ACT-R memory architecture
• ACT-R Memory vs Agent Memory
• RAG vs AI Memory
• Architecture of a memory-native organization