Extract and store facts from conversation. Strips thinking blocks. Only learns from final assistant responses and user-confirmed facts.

Retrieve relevant memories ranked by relevance, recency, and importance for the current query.

Show all memories for the current project, optionally filtered by category.

Update a specific memory fact by ID.

Clear memories by category or clear all memories for the current project.

Intelligent memory injection — returns the most relevant facts and context within a token budget.

Generate a structured checkpoint of the current session state for resuming later.

Load the most recent checkpoint (or a specific one) to resume work.

Build and query a file-to-purpose map for the project. Understands what each file does.

Estimate token usage for the current session. Warns at 70% context usage.

Move stale memories to archive. Keeps active memory lean and fast.

Surface and resolve conflicting memories. Shows both sides for user decision.

Check if a file is protected before editing. Asks for confirmation, never hard blocks.

Flag when Claude is making assumptions instead of asking. Reviews a response for unverified claims.

Review a code diff with structured explanations of each change.

Cross-check claims against known project files. Uses targeted file list only, never full repo scan.

Track and query architectural decisions with dates, reasons, and affected files.

Save and retrieve reusable prompt templates.

Full-text search across past session data.

Export session data to ./clarik-exports/ directory. Never exports to ~/Downloads.