Logo
Key Innovations

Persistent Debug Memory (PDM)

How Chronos-1 learns from past debugging sessions and builds long-term repository-specific intelligence.

Persistent Debug Memory (PDM) is Chronos’ long-term learning system. It continuously accumulates knowledge from debugging sessions, allowing Chronos to improve over time on a specific codebase.

Key Features

  • Learns from over 15 million debugging sessions
  • Improves success rate from 35% to 65% as repository-specific memory grows
  • 87% cache hit rate for recurring or similar bugs
  • Recognizes patterns across sessions
  • Learns temporal behavior and evolving project lifecycles

What PDM Stores

PDM creates a semantic memory of the entire repository, capturing:

  • Full AST snapshots and semantic embeddings per commit
  • Bug patterns, including failure signatures and stack traces
  • Fix history with validated patches
  • CI/CD logs tied to failures and resolutions
  • Documentation, comments, design notes, and PR references

How Retrieval Works

PDM combines temporal, semantic, and structural signals to retrieve the most relevant debugging context, even when an exact match does not exist.

This enables Chronos-1 to recognize new failures that resemble older ones, accelerating diagnosis and improving fix reliability.

Key Innovations

Core innovations that enable Chronos-1 to deliver state-of-the-art debugging accuracy and repository-scale reasoning.

Adaptive Graph Guided Retrieval (AGR)

Chronos’ multi-hop graph reasoning system for extracting precise debugging context at repository scale.

On this page

Key Features
What PDM Stores
How Retrieval Works