Changelog

All notable changes to m1nd are documented here. This project uses Semantic Versioning.

v0.1.0 – Initial Release

The first public release of m1nd: a neuro-symbolic connectome engine with Hebbian plasticity, spreading activation, and 43 MCP tools. Built in Rust.

Compressed Sparse Row (CSR) graph with forward and reverse adjacency
PageRank computation on ingest
4-dimensional spreading activation: structural, semantic, temporal, causal
Hebbian plasticity: Long-Term Potentiation (LTP), Long-Term Depression (LTD), homeostatic normalization
XLR differential processing: noise cancellation inspired by balanced audio cables
Hypothesis engine: claim testing with Bayesian confidence on graph paths
Counterfactual engine: module removal simulation with cascade analysis
Structural hole detection: topology-based gap analysis
Resonance analysis: standing wave computation for structural hub identification
Fingerprint engine: activation fingerprinting for structural twin detection
Trail system: investigation state persistence, resume, and multi-trail merge with conflict detection
Lock system: subgraph pinning with sub-microsecond diff (0.08us)
Temporal engine: co-change history, velocity scoring, decay functions
Domain configurations: code, music, memory, generic presets with tuned decay half-lives

Language extractors: Python, Rust, TypeScript/JavaScript, Go, Java
Generic fallback extractor: heuristic-based for unsupported languages
JSON adapter: structured data ingestion
Memory adapter: text corpus ingestion
Reference resolver: cross-file import and call resolution
Incremental ingest: re-process only changed files
Multi-repo federation: unified graph with automatic cross-repo edge detection

43 MCP tools across 7 layers:
- Foundation (13): activate, impact, missing, why, learn, drift, health, seek, scan, timeline, diverge, warmup, federate
- Perspective Navigation (12): start, routes, follow, back, peek, inspect, suggest, affinity, branch, compare, list, close
- Lock System (5): create, watch, diff, rebase, release
- Superpowers (13): hypothesize, counterfactual, predict, fingerprint, resonate, trace, validate_plan, differential, trail.save, trail.resume, trail.merge, trail.list, seek
JSON-RPC over stdio: compatible with MCP protocol version 2024-11-05
Dual transport: framed (Content-Length headers) and line-delimited JSON-RPC
Auto-persistence: configurable interval (default: every 50 queries) + on shutdown
Multi-agent support: agent ID tracking, perspective isolation, shared graph
Tool name normalization: underscores automatically converted to dots (e.g., m1nd_activate -> m1nd.activate)

Semantic scoring uses trigram matching, not neural embeddings (planned for v0.2)
No tree-sitter integration yet (planned for v0.2)
6 languages with dedicated extractors; others use generic fallback
Graph is fully in-memory; very large codebases (400K+ files) need ~80MB
No dataflow or taint analysis (out of scope; use dedicated SAST tools)