Kodezi/Chronos
Chronos
Chronos est un agent d'ingénierie logicielle LLM et un modèle de débogage à l'échelle du dépôt conçu pour la correction autonome de bugs. Le système fonctionne comme un système de correction de bugs automatisé qui localise les défauts, raisonne sur les causes profondes et implémente des correctifs multi-fichiers validés.
Le projet se distingue par un moteur de récupération guidé par graphe qui utilise un graphe de mémoire persistante pour naviguer dans les relations d'appel et les flux de données à travers de grands dépôts. Il emploie un flux de travail de débogage persistant qui indexe un historique des commits et des journaux pour reconnaître les modèles et éviter de répéter les erreurs précédentes à travers des cycles de débogage itératifs.
Le système couvre l'analyse des causes profondes via la localisation de traces causales et le traçage de bugs basé sur la logique à travers plusieurs fichiers. Il fournit en outre une réparation de code autonome en synthétisant des correctifs et en les validant au sein d'un environnement sandbox itératif. L'intelligence du dépôt est gérée via une récupération contextuelle multi-sauts et l'ingestion multi-sources des journaux et des traces.
Le système fournit une transparence sur les étapes de raisonnement et génère une documentation de débogage automatisée, incluant des évaluations des risques et des justifications pour chaque correctif appliqué.
Features
- Autonomous Coding Workflows - Analyzes software issues and implements fixes automatically using an autonomous coding and testing loop.
- Automated Software Debugging - Provides an end-to-end system for identifying root causes and implementing fixes through iterative testing and refinement.
- Autonomous Software Engineering - Functions as an LLM agent capable of autonomously localizing bugs, reasoning through root causes, and implementing patches.
- Iterative Quality Validators - Employs iterative quality validators to propose fixes and run test suites in a sandbox until a patch is verified.
- Logic-Based Bug Tracing - Traces root causes and implements fixes for bugs spanning multiple files by reasoning through codebase logic.
- Root Cause Tracing - Identifies the root cause of software issues by analyzing causal traces across a repository at scale.
- Chain-of-Cause Reasoning - Identifies software defects by synthesizing multi-source inputs and tracing causal relationships.
- Cross-File Root Cause Analysis - Identifies the underlying source of software defects by analyzing code, logs, traces, and documentation across multiple files.
- Graph-Based Context Retrieval - Implements graph-based context retrieval to navigate call relationships and dataflows across large repositories for high-precision AI retrieval.
- Graph-Based Retrieval Engines - Employs a graph-based retrieval engine and persistent memory graph to navigate complex repository dataflows.
- Repository-Scale Debugging Models - Analyzes architectural relationships, logs, and traces across entire codebases to identify and repair defects.
- Agentic Debugging Loops - Executes an iterative loop of proposing fixes and testing them in a sandbox until the code passes validation.
- Agent Memory Persistence - Maintains a graph-indexed history of commits and logs to retain lessons learned and avoid repeating mistakes.
- Call Graph Traversals - Uses high-precision call graph traversals to find relevant code segments across large-scale repositories.
- Code Context Retrieval - Retrieves relevant code segments by following execution paths and logic flow across multiple linked artifacts in a repository.
- Contextual Code Retrieval - Implements deep retrieval of relevant code snippets using a persistent memory graph to provide context for debugging.
- Refactoring Fix Validation - Tests proposed code changes in real-time against pipelines and test suites to ensure correctness before deployment.
- Multi-File Patch Synthesis - Generates complex, multi-file code fixes and supporting tests through a specialized debugging-focused model.
- Multi-File Patch Generation - Creates complex code fixes and supporting tests across a large-scale repository using a specialized debugging model.
- Execution Path Localization - Retrieves relevant code segments by following execution paths and logic flow instead of relying on filename matches.
- Global Repository Context Analysis - Performs holistic analysis of the entire codebase to identify architectural dependencies and resolve complex bugs.
- Autonomous Root Cause Synthesis - Synthesizes patches and generates tests by analyzing multi-source inputs through a specialized reasoning core.
- Patch Validation Execution - Synthesizes code fixes and validates them against a full test suite to ensure correctness before acceptance.
- Iterative Fix Loops - Executes an iterative loop of proposing code fixes and validating them in a sandbox until the issue is resolved.
- Multi-File Autonomous Repair - Applies multi-file patches by cycling through proposing fixes, running tests, and refining results until resolution.
- Autonomous Multi-File Bug Fixing - Generates and validates multi-file patches that are automatically tested against the codebase to ensure correctness.
- Persistent Debugging Memory - Maintains a memory of previous troubleshooting attempts and lessons learned to avoid repeating mistakes.
- Debugging Pattern Recognition - Recognizes patterns from previous debugging sessions to improve the success rate of solving similar issues over time.
- Graph-Indexed Debug Memory - Stores a graph-indexed history of commits and logs to retain knowledge and accelerate the resolution of recurring bugs.
- Debugging Lesson Persistence - Stores lessons learned during a session in persistent memory to avoid repeating previous mistakes.
- Multi-Source Debugging Ingestion - Ingests logs, traces, and source code from multiple data sources to understand complex software system behavior.
- Call Graph Traversal - Provides graph-guided traversal of call relationships and dataflows to analyze architectural connections across large codebases.
