All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Slack integration: AI assistant with working state feedback delivered via Slack (#14)
- Infra context repos: Collect service-scoped files and diffs from org-wide Terraform/Argo CD repos during investigations (#16)
- Token/cost tracking: Token usage and cost estimation displayed in pipeline reports and UI case screen (#11, #13)
- GitHub service layer: Git mirror cache, regression analysis, and structured service layer for GitHub evidence (#7)
- K8S_VERIFY_SSL flag: Option to skip TLS verification for EKS clusters with CA cert issues (#15)
- Infra context: Improved service name resolution for infra context lookups (#17)
- UI: Improved report readability and renamed browser tab title to Tarka (#10)
- Chat: Added component grouping to
cases.toptool (#8) - Runtime: Prevented event loop blocking in streaming chat runtimes (#5)
- Deploy: Removed External Secrets Operator dependency, simplifying deployment (#9)
- Infra: NATS retry improvements, DB query optimization, LLM observability enhancements (#6)
- Investigation pipeline: Converts Prometheus/Alertmanager alerts into structured triage reports
- Playbook system: Alert-specific investigation logic for CPU throttling, OOM kills, HTTP 5xx, pod health, job failures, and crash loops
- Diagnostic engine: Universal failure mode detection with confidence-scored hypotheses
- Deterministic base triage: Evidence-backed verdicts with explicit unknowns (never guesses)
- Multi-source evidence collection: Prometheus metrics, Kubernetes context, and log correlation
- Webhook mode: In-cluster deployment with Alertmanager webhook receiver and NATS JetStream worker pool
- Console UI: React-based case browser with investigation details and assistant chat
- Chat runtime: Tool-using conversational interface for deeper investigation (PromQL, K8s, logs, memory)
- Case-based reasoning: PostgreSQL-backed memory system for learning from past incidents
- Multi-provider LLM support: Optional enrichment via Vertex AI (Gemini) or Anthropic (Claude)
- AWS evidence collection: EC2, EBS, ELB, RDS, ECR, CloudTrail, and IAM/IRSA diagnostics
- GitHub integration: Recent commits, deployments, and PR correlation for change context
- Read-only design: All operations are strictly read-only; no cluster mutations