Skip to content

Releases: ApiliumCode/mayros

Mayros v0.2.1

14 Mar 23:56

Choose a tag to compare

Memory health tools — conflict detection and digest summaries for proactive memory maintenance.

MCP Server

  • mayros_memory_conflicts — scan for duplicate memories and graph-level contradictions (same subject+predicate, different values)
  • mayros_memory_digest — summarize memory state: total count, category distribution, recent entries, DAG stats
  • Parallel fetching in digest tool (content, categories, graph stats, DAG stats via Promise.all)
  • Both tools degrade gracefully when Cortex is down or DAG is disabled

CLI

  • mayros memory conflicts — scan Cortex for contradictions and duplicates (supports --json, --limit)
  • mayros memory digest — summarize stored memories with categories and recency (supports --json, --limit)
  • Both commands support --cortex-host, --cortex-port, --cortex-token flags

Tests

  • 13 new tests for memory health tools (duplicates, graph conflicts, empty state, Cortex down, limit capping, sort order, DAG disabled)

v0.2.0: Semantic DAG integration

13 Mar 19:42
7f9deca

Choose a tag to compare

Semantic DAG integration — full audit trail, time-travel, and verifiable history for the knowledge graph.

Semantic DAG

  • 12 CortexClient methods: dagTips, dagAction, dagHistory, dagChain, dagStats, dagPrune, dagAt, dagDiff, dagExport, dagSync, dagSyncPull, dagVerify
  • DAG enabled by default for all users — opt-out via cortex.dag.enabled = false
  • Requires AIngle Cortex >= 0.6.1

MCP Server

  • 10 new DAG tools: mayros_dag_tips, mayros_dag_action, mayros_dag_chain, mayros_dag_history, mayros_dag_time_travel, mayros_dag_diff, mayros_dag_export, mayros_dag_stats, mayros_dag_verify, mayros_dag_prune
  • 2 new DAG resources: mayros:///dag/tips, mayros:///dag/stats
  • dag-audit prompt for guided audit workflows
  • authToken now passed to all MCP tool factories (memory, cortex, DAG)
  • All tools have 30s request timeout and Cortex-offline resilience

CLI

  • mayros dag with 10 subcommands: tips, action, history, chain, stats, export, diff, at, verify, prune
  • Prune requires interactive confirmation or --yes flag
  • Default Cortex port corrected to 19090 across all 12 CLI modules
  • Client lifecycle cleanup (destroy()) in all CLI modules

Infrastructure

  • Centralized Cortex client resolution (cortex-resolution.ts)
  • postinstall binary rename for platform-suffixed GitHub Release assets
  • Version 0.1.16 → 0.2.0

Mayros v0.1.16

13 Mar 06:17
153a059

Choose a tag to compare

chore: bump to v0.1.16 and require Cortex >= 0.5.0 (#35)

## Summary
- Bump Mayros version from 0.1.15 to 0.1.16 across 55 package.json files
- Bump REQUIRED_CORTEX_VERSION from 0.4.3 to 0.5.0

## Test plan
- [x] No code changes, version bump only

v0.1.15

12 Mar 22:23
0ead86d

Choose a tag to compare

What's Changed

MCP Server

  • Built-in MCP server with 9 tools (memory, budget, governance, cortex)
  • Streamable HTTP + legacy SSE + stdio transports
  • mayros serve CLI command
  • mayros mcp-setup for Claude Code/Desktop auto-configuration

Production Hardening

  • CORS default-deny when allowedOrigins is empty
  • SSE session cap (50) to prevent connection exhaustion
  • Content-Type validation on POST endpoints
  • Timing-safe bearer token comparison
  • Command injection prevention (host/port validation, execFileSync)
  • Graceful shutdown with cached promise pattern
  • skipSignalHandlers option for managed sidecar lifecycle
  • JSON parse resilience for non-JSON Cortex responses
  • Input validation (Array.isArray for tags, limit caps)
  • Word-boundary regex in governance policy matching
  • 20 hardening tests

Infrastructure

  • Bump Node Docker image to 2e45682
  • Require Cortex >= 0.4.3

Full Changelog: v0.1.14...v0.1.15

Mayros v0.1.14

11 Mar 20:06
44db377

Choose a tag to compare

Intelligent routing, multi-agent consensus, and execution safety.

Eruberu — Adaptive Model Routing

  • Q-Learning model selector: learns optimal provider/model per task type, budget level, and time slot
  • Budget-driven fallback: auto-switches to cheaper models when budget exceeds configurable thresholds
  • Task classifier: keyword-based prompt classification (code, chat, analysis, creative)
  • Cortex persistence: Q-table stored as RDF triples with JSON file fallback
  • Integrates via before_model_resolve hook — zero changes to core execution path
  • New tools: routing_status, routing_set_strategy
  • New CLI: mayros routing status|strategy|reset

Miteru — Intelligent Task-to-Agent Routing

  • Q-Learning agent selector: learns which agent handles each task type best
  • Task classification by type, complexity, and language domain
  • Performance tracker: EMA-based agent scoring with Cortex persistence
  • Integrated into workflow orchestrator as optional routing layer
  • New tool: mesh_route_task

Kimeru — Multi-Agent Consensus

  • Three consensus strategies: majority vote, weighted (by EMA score), LLM-arbitrated
  • Automatic conflict resolution when parallel agents produce conflicting results
  • Confidence scoring and detailed vote breakdown
  • New tools: mesh_agent_performance, mesh_consensus

Tomeru — Rate Limiting & Loop Breaking

  • Sliding window rate limiter: per-tool call limits with configurable windows
  • Global token bucket: burst protection across all tools
  • Loop breaker: SHA256-based identical-call sequence detection
  • Velocity circuit breaker: hard block on runaway execution
  • Configurable modes: enforce, warn, off
  • New tools: rate_limit_status, rate_limit_adjust
  • New CLI: mayros ratelimit status|adjust|reset

Token Economy Enhancements

  • Response cache (Oboeru): LRU cache with TTL for observational response deduplication
  • Budget bridge: Symbol-based cross-plugin bridge exposes BudgetTracker to routing subsystems
  • Cache savings tracking in budget summaries

Model Router

  • buildFromPricingCatalog(): construct router from token-economy pricing catalog
  • routeWithBudget(): budget-aware routing that filters by remaining spend

Infrastructure

  • 55 extensions synced at v0.1.14
  • 55 new tests across 7 test files (Q-Learning, task classification, routing, performance tracking, consensus, rate limiting, loop breaking)
  • Auto-release workflow: GitHub Releases created automatically on version tags

v0.1.11

09 Mar 19:24
19ed343

Choose a tag to compare

Auto-update Cortex binary

  • Cortex sidecar now automatically downloads the latest binary when the installed version is older than required (0.4.1)
  • Covers all scenarios: first install (no binary), outdated binary, and fresh reinstall
  • Falls back gracefully to existing version if download fails
  • Version bump across all 49 extensions

v0.1.10

09 Mar 16:49
600e970

Choose a tag to compare

Persistent Cortex Storage & Seamless Auto-Update

  • Persistent storage: Cortex sidecar now uses --db flag for Sled-backed persistence (~/.mayros/cortex-data/)
  • Graceful update flow: flush data via /api/flush + SIGTERM before binary replacement — zero data loss
  • Sidecar hardening: lock files, port conflict detection, stderr capture, stale lock reclaim, auto-restart fix
  • Lifecycle callbacks: plugin-to-infra bridge for coordinated sidecar restarts during updates
  • Secrets migration: credentials follow dataDir with automatic migration from ~/.mayros/
  • UX fix: slash commands (/kg, /tools, /trace, etc.) now show clean user-facing text
  • Requires Cortex ≥ 0.4.1 (auto-installed if missing)

v0.1.9 — Ineru rename & Cortex 0.4.0

09 Mar 13:10

Choose a tag to compare

What's Changed

Rebranding

  • Rename titans-client.tsineru-client.ts (TitansClientIneruClient)
  • Update all imports, variables, and string references across memory-semantic and semantic-skills
  • Update docs (README, CHANGELOG, VISION)

Versioning

  • Bump REQUIRED_CORTEX_VERSION from 0.3.8 to 0.4.0
  • Bump Mayros to 0.1.9 (root + 49 extensions synced)

Aligns with AIngle v0.4.0 rebrand (titans_memory
ineru crate).