helpdesk

aiHelpDesk is an accountability layer for teams using AI agents on production databases. It diagnoses incidents, proposes remediation, and executes fixes across VMs, bare metal, Docker/Podman containers, or Kubernetes clusters. Every consequential action is approved, audited, and certified as consistent. **Core capabilities:** Fault injection testing — inject 32 known failure modes (SQL-only, SSH, K8s) against a staging database to score the agent's diagnosis and verify remediation recovery. The `faulttest` tool works without cluster access. Fleet operations — coordinate changes across multiple databases with canary phases, approval gates, schema drift detection, and full audit trails. A natural-language Planner converts free-text requests into fleet plans. Playbooks — saved runbooks combining intent with expert guidance. System playbooks ship with the tool; custom playbooks are authored, imported, or auto-generated from incident traces. AI governance — eight-module framework including tamper-proof audit logs, blast-radius enforcement, off-hours guards, and real-time policy evaluation. Human approval is required before any destructive action. Incident diagnostics — the incident agent collects database, K8s, OS, and storage layer data into a timestamped support bundle. On resolution, it automatically synthesizes a playbook draft from the audit trace. **Institutional memory:** Every resolved incident proposes a playbook draft; every successful `faulttest` auto-saves a draft. Human operators review and activate them. Free-text queries are matched against encoded symptoms and routed automatically. **Stability certification:** Before a playbook enters live rotation, it is certified across three dimensions: outcome (did it pass?), conclusion (did the agent reach the same diagnosis every run?), and evaluation (did the judge agree with itself?). Certificates like `STABLE(7) attr=oom-kill (7/7)` provide auditable proof of consistent, correct reasoning. **Architecture:** Built on Google ADK for Go. Expert agents (Database, Kubernetes, Sysadmin, Incident, Orchestrator) are swappable or extensible via the A2A protocol. The system is model-agnostic and tested with Anthropic and Gemini models. **Deployment options:** Docker/Podman, direct host installation, or Kubernetes via Helm. Binaries are available for Linux x86-64, ARM (Graviton, Ampere), and macOS (Intel and Apple Silicon). **Upstream agents:** aiHelpDesk agents can be called by humans via the Orchestrator or by upstream agents and programs via the A2A protocol and Gateway REST API. Companion bots include SREBot (observability watcher), SecBot (security responder), and GovBot (compliance reporter).