Agentdb Status

AgentDB Status

Reports the live state of the AgentDB Cognitive Container backing this session.

When to use

• User asks "what's in agentdb?", "is the memory working?", "show stats"
• Debugging recall quality — high miss rate? cache cold?
• Before / after a long session to see how much was learned
• CI smoke check before deploying an agent that depends on AgentDB

Steps

1. Call agentdb_pattern_stats to get pattern count, hit rate, recent insert/search latencies.
2. If available, call agentdb_reflexion_stats for episode count and per-task win rates.
3. Call agentdb_bandit_stats (if agentdb-learning is installed) for arm reward summaries.
4. Render a compact table:

   patterns        12,847
   episodes           384
   skills              22
   hit rate          94.2%
   avg search       0.83 ms
   learning gain   +18% since init
   storage           38 MB (.rvf)
   

5. If any number looks off (zero patterns, hit rate <50%, latency >10ms), flag it and link to the relevant ADR / docs.

Don't

• Don't dump the raw stats JSON — readers want the summary.
• Don't infer "broken" from a fresh init (zero patterns is correct on day 0).