HolmesGPT changelog: what's new each month
Every stable HolmesGPT release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 3 months covered; the current month updates daily.
September 2026
2 releases: 0.41.0 → 0.42.0
September brought security improvements, better skill management, and expanded integration capabilities. The month focused on reliability, authentication handling, and support for new diagnostic scenarios.
Highlights
- Skills synced from git now refresh in real-time without requiring restarts.
- JWT tokens are automatically refreshed before expiration to prevent authentication failures.
- New configurable timeout options for New Relic integrations.
- Added GPU node diagnostics for Kubernetes environments.
- Multiple security patches applied to address CVEs.
New
- Conversation links now included in system prompts for better context.
- GPU node diagnostics support in kubernetes-remediation toolset.
- Namespace-scoped RBAC helm value for confined Holmes installations.
- MCP structured content support with optional parameter filtering.
Improved
- Supabase API key retrieval now uses caching with local fallback for reliability.
- JWT token refresh mechanism prevents expiration-related authentication issues.
- Coralogix UI permalinks now work correctly in US2 and other regions.
- Git-synced skills use first-class skillrepos for better management.
Fixed
- Coralogix region-specific permalink generation for US2 and other regions.
- Kubernetes-remediation file-read policy defaults hardened in Helm configuration.
- MCP toolset test fixture no longer exposes project slugs.
- Null optional parameters in MCP structured content now properly dropped.
August 2026
4 releases: 0.38.1 → 0.40.0
August brought security hardening, documentation expansion, and integration updates to HolmesGPT. The month focused on fixing vulnerabilities, removing experimental features, and improving cross-platform compatibility.
- Experimental Azure SQL toolset removed
- MariaDB MCP integration removed; use native MariaDB datasource instead
- Stale Sonnet 4.0/4.5 model recommendations removed from documentation
Highlights
- Fixed multiple command injection vulnerabilities in Kubernetes and cloud toolsets
- Added GitHub App support for multiple organizations in the GitHub MCP addon
- Removed experimental Azure SQL toolset and consolidated MariaDB integration
- Enhanced OAuth flow with RFC 8707 resource indicator support for MCP
- Fixed Windows Docker build compatibility with manifest pinning
New
- GitHub App credentials support for pulling skills
- RFC 8707 resource indicator support in MCP OAuth flow
- Atlassian Rovo MCP integration documentation
- Cross-cluster tools documentation
- Bitbucket instructions for loading custom skills
- LLM eval for AKS --query command injection detection
Improved
- kubectl-run toolset hardened to run commands without a host shell
- Tool memory limit logging when OOM killed
- GitHub MCP addon now supports GitHub App installed on multiple organizations
- Docker build manifest handling for Windows clones
Fixed
- Command injection vulnerability in Kubernetes toolset
- Command injection vulnerabilities in slab, kubevela, inspektorgadget, and aks toolsets
- SSRF vulnerability in internet/fetchwebpage and connectivitycheck toolsets
- Operator image startup by removing holmes package dependency
- Behavior-guard permissions evaluation setup
July 2026
4 releases: 0.35.0 → 0.38.0
July brought performance improvements, enhanced security patching, and expanded integration capabilities to HolmesGPT. The month focused on operational visibility, reducing token usage, and strengthening authentication options.
- Removed KRR resource recommendation features.
Highlights
- Reduced system prompt and tool description sizes by 50-60%, lowering token consumption.
- Added remote tool call approval workflows for better operational control.
- Implemented new /api/info endpoint for operational visibility and monitoring.
- Extended support for Langfuse in OpenTelemetry tracing and pre-acquired Azure tokens.
- Made conversation-worker slot exhaustion visible to help diagnose capacity issues.
New
- JSON logging format support for log scrapers.
- Langfuse integration with OpenTelemetry tracing.
- /api/info endpoint for operational visibility.
- Remote tool call approval workflows.
- Support for pre-acquired AZUREADTOKEN for Entra ID authentication.
- Kubernetes MCP OAuth information added to data sources.
- Multi-instance-capable data source marking in catalog.
- Slack channel-history context recovery evaluations.
Improved
- Reduced system prompt and tool description sizes by approximately 50-60% to lower token usage.
- Switched Holmes image to Alpine base for smaller container footprint.
- Bound conversation and tool-call claiming to worker capacity.
- GitHub MCP PAT path now uses native HTTP mode.
- Consolidated Datadog and Elasticsearch/OpenSearch data source catalog cards.
- Made conversation-worker slot exhaustion visible and bounded reconnect sign-in.
- Clarified namespace requirements for Holmes service account setup.
- Updated requestsource field documentation with taxonomy cross-references.
Fixed
- Security patch: bumped litellm to 1.89.0 to address CVE-2026-49468.
- Fixed Holmes operator health checks when using VertexAI/Gemini.
- Fixed wrong icon display for SQL Server and other database subtypes.
- Fixed log filter leaks in parallel test execution.
- Corrected tampered-signature test validation.
Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. HolmesGPT: pricing, features and alternatives · All changelogs