New features
Integrations
- Argo CD investigations — HealOps can now investigate Argo CD deployment issues, surfacing sync status and rollout evidence directly in incidents. See Argo CD.
- Apache Airflow — New integration ingests DAG and task evidence so HealOps can reason about pipeline failures during investigations. See Airflow.
- VictoriaLogs — Query VictoriaLogs as a log source during investigations. See VictoriaLogs.
- Amazon RDS — Native RDS integration adds describe-instance and events tools, plus richer Performance Insights evidence in root-cause analysis. See RDS.
LLM providers and CLI backends
HealOps now supports four additional CLI-based LLM backends, so you can run investigations through whichever agent CLI you already use:- Claude Code CLI as an LLM provider
- Cursor Agent CLI as a non-interactive LLM backend
- opencode CLI as an LLM provider
- Kimi Code CLI as a non-interactive LLM backend
- Requesty added as a hosted LLM provider
Interactive shell
- Documentation-aware interactive shell — The interactive shell can now answer questions using HealOps docs as context, so you can ask product questions without leaving the terminal.
- Interactive REPL — First release of a Claude-Code-style SRE terminal for running investigations interactively.
- UX improvements — General polish to interactive shell ergonomics and automation flows.
Alerts and notifications
- Canonical alert payloads — Incoming alerts from any source are now normalized to a single HealOps schema, making routing and triage more consistent across providers.
- Release announcements — New workflow posts HealOps releases to Discord and X automatically.
Updates
- Toolcall model switching at runtime — Swap the model used for tool calls mid-session without restarting the CLI.
- Onboarding hint after install — The CLI now prints a clear next-step hint after a successful install.
- Improved RDS storage-full RCA — Better evidence handling for storage-exhaustion scenarios.
- Tracer migrated to opensre — The legacy tracer path is deprecated; functionality has moved into the main code path.
- Analytics opt-out —
HEALOPS_NO_TELEMETRYis now honored consistently across the CLI.
Bug fixes
- Discord and Slack delivery — Hardened against non-JSON failures and accidental token leakage in error paths.
- Bedrock support — Universal Bedrock model support via the Converse API.
- Claude Code onboarding — Live macOS Keychain probe and more reliable auth status parsing during onboarding.
- Codex and Claude Code auth probes — Hardened binary resolution and auth probe paths for CLI-based providers.
healops tests— FixedFileNotFoundErrorwhen running in the bundled binary.- Interactive shell — Fixed Ctrl+C double-exit behavior.
- Health-check noise — Health checks with
state=normalare no longer incorrectly flagged as noise. - Guardrails — Overlapping redaction spans are now merged so sensitive bookends can’t leak.
- WAL-driven replication lag RCA — More accurate root-cause output for the PostgreSQL replication-lag scenario.
- Performance Insights evidence — Bad-query evidence is now surfaced in RCA output.
- CLI onboarding output — Removed an extraneous summary block from
onboardcompletion. - Unique CLI install IDs — Analytics now persist a stable install ID per CLI.
Tracer