AgentDB v3 (3.0.0-alpha.11) is a proof-gated graph intelligence database for AI agents, requiring cryptographic proof for every state mutation. It provides native Rust performance via @ruvector/graph-transformer with 8 verified graph modules, including HNSW search, causal reasoning, and reflexion memory. The library supports episodic memory, skill libraries, and lifelong learning with explainable provenance. Released as alpha, it targets Node.js >=18.0.0 and ships TypeScript types. Key differentiators: mandatory proof verification on mutations, causal reasoning capabilities, and native Rust acceleration.
npm install agentdbVerified import paths — ran on the pinned version, not inferred.
Creates an AgentDB instance, initializes a memory graph with 128 dimensions, inserts a node, generates a cryptographic proof, and performs a search query.
Update all code to use the new v3 API. Refer to migration guide at https://github.com/ruvnet/agentic-flow/wiki/v3-migration.
Replace Agent with AgentDB.create() and MemoryGraph with initGraph(). Use subpath imports for specific modules.
Always call .prove() before mutation or use the proof returned from the mutation method if available.
Upgrade Node.js to 18 or later. If stuck on older Node, use agentdb v2.x (last v2 release is 2.3.6).
Change to import { AgentDB } from 'agentdb' and ensure your project uses ES modules (set 'type': 'module' in package.json).Before mutating, call graph.prove() with the appropriate parameters and pass the proof to the mutation method.
If using TypeScript, ensure 'moduleResolution' is set to 'node16' or 'bundler'. For JavaScript, check that the package is correctly installed.