Quazzar Intelligence

Quazzar Intelligence is the umbrella name for three capabilities that make your node’s AI assistant trustworthy, well-informed, and proactive. They work independently and together — and they all run entirely on your node, with no data sent to any external service.

Verified Answers

When Molly answers a question that involves numbers, metrics, dates, or names, every figure is checked against the real data Molly retrieved for that turn. If a figure can’t be traced back to a real source, it is marked with ⚠️ inline — so you always know what’s solid and what needs a second look.

What this means for you:

• Confident answers carry a Sources footer showing where each value came from.
• Unverifiable figures are flagged visibly instead of being silently invented.
• If Molly doesn’t have the data to answer reliably, she says so rather than guessing.

This applies to Molly chat, Molly’s proactive diagnostics, and the Control Panel AI chat.

Environment control: on by default. Set QUAZZAR_AI_INTEGRITY=off to disable (not recommended for normal use).

→ See Proactive Molly for how Molly uses verified answers in proactive alerts.

Smart Memory

Your notes and saved knowledge are searched with a ranking engine that combines exact-word matching with meaning-based similarity. Whether you remember the exact phrase or just the idea, the right notes surface.

What this means for you:

• Finds related notes together — notes linked to a strong match appear alongside it, so you get the full cluster of relevant context, not just a single result.
• Understands synonyms automatically — “k8s” finds “Kubernetes”, “MRR” finds “monthly recurring revenue”. No manual tagging needed.
• Keeps your memory tidy — near-identical notes are merged automatically so duplicates don’t pile up over time.
• Stemming — “deploying”, “deployed”, “deploy” all find the same notes.

This powers both Molly’s own recall and the Memory MCP tool used by Claude Desktop, Claude Code, Cursor, and other MCP clients.

Environment control: the improved ranking ships in shadow mode by default (it computes both rankings in the background for validation). Set MEMORY_SEARCH_MODE=hybrid_v2 to activate it as your live ranking.

→ See Molly Memory and Memory MCP for full details.

Predictive Insights

Your node runs four on-device health predictors every hour and writes calibrated results to a local database. You can ask Molly about them, query them from any MCP client, or glance at the dashboard card — and if you opt in, Molly will proactively alert you when something looks risky.

What’s monitored:

• Disk capacity forecast — projects when your disk will fill up based on your actual usage trend (“disk projected full in ~9 days”).
• Resource anomaly — flags CPU, RAM, disk I/O, or network readings that are unusually high or low for this time of day and day of week.
• Backup health — watches for rising failure rates or consecutive backup failures before they become a problem.
• App health — tracks restart storms, OOM kills, and crash events per app.

Two things that keep it useful:

• Honest refusal — a fresh node with little history returns “not enough data yet” rather than an invented forecast. Predictions only appear once there’s real signal.
• Quiet by default — results are always available on demand (Molly, MCP, REST, dashboard card), but Molly only proactively notifies you when you opt in.

Environment control: detection and on-demand serving are on by default. Set QUAZZAR_INSIGHTS_ALERTS=true to enable proactive Molly notifications for high-risk predictions.

→ See Predictive Insights for full details on predictors, the REST endpoint, and the MCP tool.

Configuration at a glance