AgentDB Advanced Features
AgentDB is a standalone RuvNet package, not this deployment's memory path; durable agent/project memory here is RuVector-postgres via mcp__claude-flow__memory_* (bge-small-en-v1.5, 384-dim).
Distributed and advanced AgentDB patterns. For AgentDB architecture, performance
benchmarks, and common API patterns, see AgentDB Overview.
When to use
Reach for this skill when a plain AgentDB setup no longer covers the need:
- Distributed sync — keep patterns consistent across AgentDB instances on
different hosts (QUIC).
- Multiple / sharded databases — route or scale across separate
.db files.
- Advanced retrieval — custom distance metrics, hybrid vector+metadata
filtering, weighted scoring, MMR diversity, or synthesized context.
- Reinforcement learning — build self-improving agents that train on logged
experience (9 RL algorithms).
When not to use
- Basic vector search or single-database setups →
agentdb-vector-search.
- Simple agent memory (session, long-term) →
agentdb-memory-patterns.
- Performance tuning without distributed features →
agentdb-vector-search.
- Non-AgentDB vector databases (pgvector, Pinecone, Weaviate) — this skill is
AgentDB-specific.
Prerequisites: distributed-systems basics (for QUIC sync) and vector-search
fundamentals.
Quick path
import { createAgentDBAdapter } from 'agentic-flow/reasoningbank';
// Distributed adapter with QUIC sync + hybrid retrieval
const adapter = await createAgentDBAdapter({
dbPath: '.agentdb/distributed.db',
enableQUICSync: true,
syncPort: 4433,
syncPeers: ['192.168.1.11:4433', '192.168.1.12:4433'],
});
const result = await adapter.retrieveWithReasoning(queryEmbedding, {
metric: 'cosine', // or 'euclidean' | 'dot'
k: 20,
useMMR: true, // diverse results
filters: { year: { $gte: 2023 } }, // hybrid metadata filter
});
Reference tiers (load on demand)
| Topic |
File |
| QUIC synchronization — enable, config, multi-node deploy, env vars, troubleshooting |
references/quic-sync.md |
| Search features — distance metrics, hybrid/weighted search, MMR, context synthesis |
references/search-features.md |
| Deployment — multi-database, sharding, connection pooling, error handling, CLI import/export/optimise |
references/deployment.md |
| Reinforcement-learning plugins — 9 algorithms, training API, decision-transformer config |
references/reinforcement-learning.md |
Learn more
Category: Advanced / Distributed Systems · Difficulty: Advanced ·
Estimated Time: 45-60 minutes
1---2name: agentdb-advanced3description: Advanced AgentDB beyond single-database vector search: distributed QUIC sync across nodes, multi-database coordination and sharding, custom distance metrics, hybrid vector+metadata search, MMR diversity, context synthesis, and reinforcement-learning plugins. Use when an AgentDB deployment needs cross-node sync, cross-database routing, filtered/weighted hybrid retrieval, or self-improving RL agents.4---56# AgentDB Advanced Features78AgentDB is a standalone RuvNet package, not this deployment's memory path; durable agent/project memory here is RuVector-postgres via `mcp__claude-flow__memory_*` (bge-small-en-v1.5, 384-dim).910Distributed and advanced AgentDB patterns. For AgentDB architecture, performance11benchmarks, and common API patterns, see [AgentDB Overview](./docs/agentdb-overview.md).1213## When to use1415Reach for this skill when a plain AgentDB setup no longer covers the need:1617- **Distributed sync** — keep patterns consistent across AgentDB instances on18 different hosts (QUIC).19- **Multiple / sharded databases** — route or scale across separate `.db` files.20- **Advanced retrieval** — custom distance metrics, hybrid vector+metadata21 filtering, weighted scoring, MMR diversity, or synthesized context.22- **Reinforcement learning** — build self-improving agents that train on logged23 experience (9 RL algorithms).2425## When not to use2627- Basic vector search or single-database setups → `agentdb-vector-search`.28- Simple agent memory (session, long-term) → `agentdb-memory-patterns`.29- Performance tuning without distributed features → `agentdb-vector-search`.30- Non-AgentDB vector databases (pgvector, Pinecone, Weaviate) — this skill is31 AgentDB-specific.3233**Prerequisites**: distributed-systems basics (for QUIC sync) and vector-search34fundamentals.3536## Quick path3738```typescript39import { createAgentDBAdapter } from 'agentic-flow/reasoningbank';4041// Distributed adapter with QUIC sync + hybrid retrieval42const adapter = await createAgentDBAdapter({43 dbPath: '.agentdb/distributed.db',44 enableQUICSync: true,45 syncPort: 4433,46 syncPeers: ['192.168.1.11:4433', '192.168.1.12:4433'],47});4849const result = await adapter.retrieveWithReasoning(queryEmbedding, {50 metric: 'cosine', // or 'euclidean' | 'dot'51 k: 20,52 useMMR: true, // diverse results53 filters: { year: { $gte: 2023 } }, // hybrid metadata filter54});55```5657## Reference tiers (load on demand)5859| Topic | File |60|-------|------|61| QUIC synchronization — enable, config, multi-node deploy, env vars, troubleshooting | [references/quic-sync.md](./references/quic-sync.md) |62| Search features — distance metrics, hybrid/weighted search, MMR, context synthesis | [references/search-features.md](./references/search-features.md) |63| Deployment — multi-database, sharding, connection pooling, error handling, CLI import/export/optimise | [references/deployment.md](./references/deployment.md) |64| Reinforcement-learning plugins — 9 algorithms, training API, decision-transformer config | [references/reinforcement-learning.md](./references/reinforcement-learning.md) |6566## Learn more6768- **QUIC Protocol**: [references/quic-sync.md](./references/quic-sync.md)69- **Hybrid Search**: [references/search-features.md](./references/search-features.md)70- See [AgentDB Overview](./docs/agentdb-overview.md#links) for general links.7172---7374**Category**: Advanced / Distributed Systems · **Difficulty**: Advanced ·75**Estimated Time**: 45-60 minutes
dreamlab-ai