Memory

What the Agent Remembers

Currently, Burg AI has limited memory:

• Within a PR: Comments and context from the current PR
• Historical context: Limited to the same PR's discussion

What It Does NOT Remember

• Previous PRs in the same repository
• Your feedback on past reviews
• Patterns from resolved issues
• Cross-repository context

How Memory Improves Future Reviews

Even with limited memory, reviews improve through:

1. Framework detection — The agent detects your tech stack from the current PR
2. PR context — Existing PR comments are considered
3. File relationships — Related files in the same PR are analyzed together

Future: Enhanced Memory

Planned memory features (post-MVP):

• Cross-PR pattern learning
• Repository-specific style learning
• Feedback-driven improvements
• Issue deduplication across PRs

Current Best Practice

Since memory is limited:

• Use custom prompts to encode your preferences
• Add repository-specific instructions (Pro)
• Provide feedback on reviews to help us improve the system