RIFT is a local-first, forensic incident investigation and observability platform engineered specifically for autonomous AI software agents.
When modern AI coding agents (Devin, Claude Code, Cursor Composer, AutoGPT, SWE-agent) interact with filesystems, terminals, databases, and developer tools, their failures rarely resemble traditional software exceptions. Agents can delete vital source files while assuming they are temporary artifacts, enter infinite retry loops, hallucinate success despite failing builds, or silently trigger catastrophic regressions across downstream codebases.
RIFT bridges the gap between agent intent and runtime reality through a unified engineering pipeline:
High-Density Execution Graph & Timeline: Reconstructs execution traces into an interactive Directed Acyclic Graph (DAG) with tri-directional synchronized inspection.
Deterministic Incident Detection: High-speed, rule-based detector (8 core rules) operating with 0% false positives and zero LLM hallucination risk.
Google Gemma Forensic Investigation: Local-first inference using Google Gemma models via Ollama to generate multi-hypothesis causal post-mortem with 100% verified trace citations.
AST Codebase Knowledge & Blast Radius (Graphify): Extracts code syntax trees and call hierarchies with zero external API costs to map the exact blast radius of destructive agent actions.
Counterfactual Replay Lab: Containerized, dual-isolated sandbox executing typed filesystem mutations to reproduce failures and cryptographically verify developer fixes.
Executive Incident Post-Mortems: Single-click export of engineering post-mortems in GitHub-Flavored Markdown and machine-readable JSON.
Check the codebase here - https://github.com/CoderOggy78/RIFT/tree/main
Top comments (0)