AI Operations
Context-aware AI with guardrails. Reads logs, configs, and metrics, then suggests scoped actions. Changes never run without your approval, and AI tool calls are recorded in the audit log.
Supported Providers
Anthropic Claude
API key and model selection
OpenAI
GPT-4o and other models
Google Gemini
API key and model selection
Ollama
Self-hosted, no API key — runs fully offline alongside the ServerMind backend
AI Tool Capabilities
Read-Only
Auto-approved — no confirmation needed
- Browse the filesystem, search and read files
- Tail logs with regex filtering, log analytics
- Service status, dependencies and logs
- System info, metrics history, anomaly detection and resource forecasting
- Alert history, incident timeline, snapshot diffs
- Security findings and compliance status
- Read-only views of databases, web servers, runtimes and queues
Mutating
Single approval required
- Write files via SafeApply (snapshot first)
- Start/restart/reload services
- Create filesystem snapshots
- Install/update packages; manage users, groups and cron jobs
- Run database commands (PostgreSQL, MySQL, Redis, MongoDB, Cassandra)
- Reload or SafeApply Nginx, Apache and HAProxy config
Destructive
Approval + typed hostname confirmation
- Restore from snapshots
- Execute commands (binary + arguments, no shell)
- System restart and shutdown
- Stop services
- Remove or roll back packages, delete users
Approval Workflow & Safety
Approval enforced server-side — the tool call waits until you approve
Risk level shown on every approval request
Typed hostname confirmation for destructive operations
Tools filtered by the user's own RBAC permissions
Secret paths blocked and tool output redacted
AI tool calls, approvals and rejections recorded in the audit log
Configurable: provider, model, API key, operating mode, custom instructions
AI suggests — you decide. Read-only lookups run automatically; every change waits for your approval. No black-box changes.