Self-tuning Valkey for AI agents
LLM call caching for AI agents. Historical slowlog, anomaly detection, and pattern analysis for ops. Debug what happened at 3am, not just what's happening now.
For AI
npm install @betterdb/agent-cache iovalkey
TypeScript. Multi-tier LLM, tool, and session caching. OTel + Prometheus built in.
npm install @betterdb/semantic-cache iovalkey
TypeScript. Similarity-based caching with valkey-search.
For Ops
claude mcp add betterdb -- \ npx @betterdb/mcp betterdb-mcp \ --autostart --persist
Adds BetterDB as an MCP server to Claude Code. Starts monitoring automatically.
Then ask Claude about your database health, slow queries, or memory usage.
The cache tunes itself
No other Valkey or Redis cache library does this.
Agent MCP call
// Agent MCP call
await mcp.callTool(
'cache_propose_threshold_adjust',
{
cache_name: 'prod-semantic',
new_threshold: 0.075,
reasoning:
'hit rate 28% over 7d,'
+ ' tighten threshold',
}
)Pending proposal
// Pending proposal (API response)
{
"id": "prop_01jwx3krq5",
"status": "pending",
"cache_name": "production-semantic",
"new_threshold": 0.075,
"expires_at": "2026-05-06T12:00:00Z",
"warnings": []
}After approval
// Dispatcher writes to Valkey: HSET production-semantic:__config \ threshold 0.075 // Library picks up the change // within seconds. No restart.
The agent observes live cache metrics via MCP read tools, proposes a config change with reasoning, and a human approves it in BetterDB Monitor. The cache library polls its config key in Valkey and swaps the policy atomically - no restart, no redeploy. Config polling is live in @betterdb/semantic-cache@0.4.0 and @betterdb/agent-cache@0.6.0. See the full closed-loop example
Why teams choose BetterDB for agent caching
Three cache tiers behind one Valkey connection. No modules required.
| Capability | @betterdb/agent-cache | LangChain RedisCache | LangGraph checkpoint-redis | AutoGen RedisStore | LiteLLM Redis | Upstash + Vercel AI SDK |
|---|---|---|---|---|---|---|
| Agent-tunable via MCP | ||||||
| Live config updates (no restart) | ||||||
| Multi-tier (LLM + Tool + State) | LLM only | State only | LLM only | LLM only | LLM only | |
| Built-in OTel + Prometheus | Partial | |||||
| No modules required | Redis 8 + modules | Upstash only | ||||
| Base SDK support (OpenAI, Anthropic) | ||||||
| Multi-modal (images, audio, files) | ||||||
| Language support | TypeScript + Python | TS only | TS only | Python only | Python only | TS only |
| Framework adapters | OpenAI, Anthropic, LangChain, LangGraph, LlamaIndex, Vercel | LC only | LG only | AutoGen only | LiteLLM only | AI SDK only |
| Zero-config cost tracking | Bundled LiteLLM table, 1,900+ models |
AI Caching
Two caching strategies, one Valkey instance, full observability.
Exact-match caching
Three cache tiers: LLM responses, tool results, and session state. TypeScript and Python. Works with OpenAI, Anthropic, LangChain, LangGraph, LlamaIndex, and (TS) Vercel AI SDK.
npm install @betterdb/agent-cache iovalkey
pip install betterdb-agent-cache
Semantic caching
Similarity-based response caching with adapter parity for OpenAI, Anthropic, LangChain, LangGraph, LlamaIndex, and Vercel AI SDK. Bundled cost tracking, auto-tuning threshold recommendations, and embedding cache. TypeScript and Python.
npm install @betterdb/semantic-cache iovalkey
pip install betterdb-semantic-cache
See what's actually happening inside your Valkey or Redis instance
Historical data that survives a log rotation. Client-level attribution. Automatic anomaly detection.
500 slow queries, 114 unique patterns, 19 command types - across any historical time range. The evidence that used to disappear after a log rotation.
Built for humans and agents alike
Investigate incidents yourself - or point your agent at BetterDB and walk away.
Your agent investigates and proposes
Connect BetterDB's MCP server to Claude Code, Cursor, or any MCP-compatible client. The agent reads live cache state - hit rate, similarity distribution, per-tool effectiveness - and can propose threshold tightenings, TTL adjustments, and targeted invalidations. Humans approve; changes go live within seconds.
{
"mcpServers": {
"betterdb": {
"command": "npx",
"args": ["-y", "@betterdb/mcp"]
}
}
}You oversee and approve
Full dashboard, slowlog pattern analysis, anomaly detection, client attribution, vector search, and vector workload health. Review and approve agent-proposed cache optimizations in the BetterDB Monitor proposal queue - you see the reasoning, the expected outcome, and can edit before approving.
Everything you need to understand your Valkey or Redis instance
Deep insights into your Valkey or Redis deployment with minimal overhead.
Anomaly Detection
Automatic detection of unusual patterns across memory, CPU, and connections.
Historical Analytics
Query what happened at 3am, not just what's happening now.
Cluster Visualization
Interactive topology graphs, slot heatmaps, and migration tracking.
ACL Audit Trail
Track who accessed what for compliance and debugging.
Slowlog & COMMANDLOG
Pattern analysis across persisted slow queries and full command history. Valkey 8.1+ COMMANDLOG captures every command, not just the slowest ones.
Key Analytics
Analyze key patterns by namespace and key type.
Client Analytics
See which clients consume resources. Detect unusual buffer sizes and connection spikes.
Prometheus & Webhooks
107 Prometheus metrics plus Slack, email, and webhook notifications.
Vector Workload Analytics
FT.SEARCH ops/sec and latency over time, per-index health with alerts for indexing failures, backfill progress, and deleted-doc growth. Built for teams running RAG and semantic search in production.
Vector Search
Inspect vector indexes, browse embeddings, and find semantically similar entries with similarity scores.
Cache Proposals
Proposal queue and audit trail for agent-submitted cache optimizations. Review, edit, and approve threshold and TTL changes directly in BetterDB Monitor.
MCP Server
Full observability and cache intelligence tools from any MCP-compatible client - Claude Code, Cursor, IDEs. Agents read cache state and propose optimizations directly.
View on registry →Inference Pipeline Latency
Per-operation P50/P95/P99 profiles for FT.SEARCH, reads, and writes. Per-index SLA thresholds with webhook breach alerts - know before users do.
Hot Key Tracking
Top-50 keys by access frequency with rank movement over time. Spot what's climbing before it becomes a bottleneck.
Latency Monitoring
Per-event latency history across P50/P95/P99. Know when your instance started slowing down, not just that it is.
CPU & I/O Threads
Per-thread I/O metrics and CPU utilization charts, including Valkey 8.x I/O thread breakdowns.
Migration
Analyze compatibility, move data, and validate results across Redis, Valkey, cloud, and self-hosted - in any direction.
Throughput Forecasting
Growth rate trend and ceiling-based ops/sec forecasting. Get alerted before you hit capacity, not after.
Our goal: zero hacks for Valkey or Redis
If you've ever written a custom script, stitched together three tools, or just accepted that something wasn't observable - we want to hear about it. Tell us what's missing. We'll build it, and the next person won't have to figure it out themselves.
Book 15 min with KristiyanPrefer email? kristiyan@betterdb.com
Free during early access - every Pro and Enterprise feature is unlocked while we're in beta.
Ready to get started?
Start monitoring in minutes - no infrastructure to maintain. Team collaboration, agent-based monitoring for private databases, and more. Or self-host - open source core, zero lock-in.