agentdb

Walk the causal graph in AgentDB to explain why two memories are connected, or trace a root cause. Use when the user asks "why did X happen", "what led to Y", or after an incident.

Record a causal relationship between two memories in AgentDB — "X caused Y", "A supersedes B", "patch-foo depends-on patch-bar". Use when the user is documenting cause/effect, dependencies, supersessions, or after-action analysis.

Store memories in tier-aware hierarchical memory — working / short-term / long-term — and recall with tier filters. Use for working-set context that should fade, vs facts that should persist, vs patterns that should be searchable forever.

Initialize an AgentDB Cognitive Container (.rvf file) in the current project. Sets up storage, embedder config, and the agentdb MCP server. Use when the user is starting a new project that needs vector memory, or asks to "set up agentdb" / "init agentdb".

Show AgentDB health — pattern count, embedder status, cache hit rate, learning gain since init. Use when the user asks "is agentdb working?", "how many memories?", "show agentdb stats", or after long-running sessions to confirm state.

Execute Cypher queries against AgentDB's graph backend. Use when the user wants to write a custom traversal that the standard tools don't cover, or when explaining graph state.

Search and manage hyperedges — n-ary relationships between memories. Use for swarm membership, multi-cause incidents, or any "this involves all of (A, B, C, D)" relationship that doesn't fit a binary edge.

K-hop traversal from a starting node in AgentDB's graph. Use to explore neighborhoods, find reachable nodes, or visualize a memory's "context".

Close the learning loop — record reward signal for an action AgentDB suggested. Use after using anything from agentdb_pattern_search / reflexion_recall / skill_search / learning_route. The bandit needs the signal to improve.

Train one of AgentDB's 9 RL algorithms on a stream of episodes. Use when the user has accumulated successful/failed episodes and wants to derive a policy, or when a task type is repeated enough to benefit from RL routing.

Ask the AgentDB bandit which RL algorithm / skill / pattern fits the current task best. Use at task start when there are multiple plausible approaches and you want the data-driven pick.

Retrieve relevant memories for the current task from AgentDB. Use at the start of a task to load prior knowledge, when stuck to surface what worked before, or when the user asks "what do we know about X" / "have we done this before?"

Store a memory in AgentDB — an episode (task + outcome + critique), a pattern, or a skill. Use when the user says "remember this", "save this for later", "add to memory", or when the agent has just succeeded/failed at a task and the lesson is worth keeping.

Promote a validated pattern into a reusable Skill in AgentDB's skill library. Use when the same approach has worked 3+ times across episodes, or when the user explicitly says "make this a skill" / "save this as reusable".

Search with feature attributions — return WHY each match scored where it did. Use when debugging recall quality, auditing for bias, or explaining results to a user.

Hybrid search — BM25 keyword + dense vector fused with Reciprocal Rank Fusion. Use when queries have specific identifiers, code symbols, or proper nouns that pure semantic search might miss.

Maximal Marginal Relevance rerank — get diverse top-k instead of redundant top-k. Use when standard search returns 5 near-duplicates, or when you want broader coverage of a topic.