Tarsy-bot is a deprecated Python-based Site Reliability Engineering agent designed to automatically process alerts through intelligent chain-based workflows. The project has been superseded by a new Go and Python hybrid implementation available at the codeready-toolchain/tarsy repository, which the maintainers recommend for all new deployments and contributions. The original tarsy-bot repository represents an earlier iteration of the TARSy (Thoughtful Alert Response System) concept, which aimed to provide automated incident response and analysis capabilities for SRE teams.
The core functionality of tarsy-bot centers on accepting arbitrary JSON alert payloads from monitoring systems and routing them through sequential stages of specialized agents. These agents work in coordinated chains, with each stage building upon the analysis from previous stages. The system integrates with MCP (Model Context Protocol) servers to gather system information and execute domain-specific tools, enabling comprehensive multi-stage incident analysis without requiring code modifications to add new capabilities. Configuration-based agent deployment allows teams to define new agents and chain definitions through YAML files rather than code changes.
The architecture supports parallel agent execution, allowing multiple agents to investigate different aspects of an incident concurrently with automatic synthesis of their findings into unified analysis. The system can operate in multiple parallelism modes including multi-agent parallelism for independent domain investigation, replica parallelism for redundancy, and comparison parallelism for A/B testing different LLM providers or strategies. Multi-LLM provider support enables configuration and switching between OpenAI, Google, Anthropic, xAI, and other providers through YAML configuration, with optional Google Search grounding for Gemini models and native thinking mode for Gemini 2.0 and later versions.
Tarsy-bot includes optional GitHub runbook integration that automatically retrieves and includes relevant runbooks from GitHub repositories per agent chain, contextualizing investigations with team knowledge. The system maintains a comprehensive audit trail with stage-level timeline reconstruction for complete visibility into chain processing workflows. A web-based SRE dashboard provides real-time monitoring with live LLM streaming and interactive chain timeline visualization, while follow-up chat functionality allows engineers to continue investigating after sessions complete by asking clarifying questions or requesting deeper analysis with full context and tool access.
The platform implements pause and resume capabilities for long-running investigations, automatically pausing at iteration limits while preserving full conversation state and allowing resumption with a single click. Alternatively, automatic conclusion can be configured at iteration limits instead of pausing, with agents generating final summaries. Data masking combines code-based structural analysis for Kubernetes Secrets with regex patterns to protect sensitive information including API keys, passwords, certificates, emails, and SSH keys in MCP responses and alert payloads. Tool result summarization uses LLM-powered analysis to reduce token usage and improve agent reasoning by focusing on relevant information while preserving full results in audit logs.
According to GitGenius activity tracking, the repository shows median issue and pull request response latency of 1113.4 hours with a mean of 1524.9 hours across 11 tracked items. The most active contributor tracked was alexeykazakov with 17 events, followed by AryanPandey72 and MatousJobanek with one event each. The repository was classified into multiple categories including automation, chatbot, pull request management, issue tracking, repository maintenance, bot integration, workflow automation, comment handling, collaboration, and developer productivity. The project requires Python 3.13 or later for development mode and supports both direct backend development and containerized deployment with PostgreSQL database, OAuth2 authentication, and Nginx reverse proxy for production-like testing environments.
