Skip to main content
The memory system stores contextual knowledge about your project — decisions you’ve made, patterns you’ve discovered, and progress milestones. These memories persist across sessions and are accessible to AI coding tools via MCP.

Memory categories

Each memory belongs to a category:
CategoryUse for
DecisionArchitecture choices, technology selections, trade-off resolutions
LearningPatterns discovered, gotchas encountered, things that worked
ProgressMilestones reached, features completed, status updates
PreferenceCoding conventions, style preferences, team agreements

Memory status

Memories have a lifecycle status:
  • Active — current and relevant
  • Superseded — replaced by a newer memory
  • Conflicting — contradicts another memory (needs resolution)
  • Archived — no longer relevant but preserved for history
Memories are embedded as 1536-dimension vectors using pgvector. When you or an AI tool searches memory, Kommit uses cosine similarity to find the most relevant results — not just keyword matches.

Conflict detection

When a new memory contradicts an existing one, Kommit flags it as a conflict. You can resolve conflicts by:
  • Keeping the existing memory
  • Keeping the new memory
  • Merging both into a single memory
  • Archiving both and writing a fresh one
Search memories across all projects in your organization from the global memory page. This is useful when patterns or decisions in one project apply to another.

Memory graph

The graph view visualizes relationships between memories as a force-directed graph, showing how decisions connect to learnings and progress.

MCP integration

AI coding tools access memories through the MCP search_memory tool. When a developer asks their AI assistant about project context, it queries Kommit’s memory store and returns relevant decisions and learnings. See MCP setup for configuration.