Skip to content

CHANGELOG.md

Latest commit

 

History

History
42 lines (31 loc) · 2.33 KB

CHANGELOG.md

File metadata and controls

42 lines (31 loc) · 2.33 KB
Raw

Changelog

All notable changes to Eternal Nagual. This project is a single continuously-running organism, so "releases" are moments its keeper froze a known-good state — the agent itself never stops between them.

The format is loosely based on Keep a Changelog. Dates are UTC.

[Unreleased]

Added

  • ROADMAP.md — where the organism is heading (now / next / later / north stars)
  • COMPARISON.md — honest Nagual vs. AutoGPT / CrewAI / LangGraph / BabyAGI
  • Router map injected into the orchestrator's own system prompt — the mind now knows what substrate it thinks with
  • Self-healing infrastructure loop: core, dashboard, API health, cron loops and session all auto-recover; the keeper is paged only on a real outage
  • Autonomous following — the agent reads a bio and decides for itself whether to follow

Changed

  • Router rebuilt around interleaved free-tier slots (deepseek-v4-pro / gpt-oss-120b reasoning / glm / gemini) with an OpenRouter :free catch-all tail, so a provider-wide rate-limit storm no longer stalls the loops
  • Voice path re-ordered so fast models answer first (sub-5s), reasoning models stay as fallback

Fixed

  • Rate-limit storm that silently froze karma actions (empty-except swallowing "all slots failed")
  • False "login lost" trips on slow page loads (now retries before declaring a real cookie expiry)
  • Living-world chronicle text overflow and page scroll

[2.0.0] — 2026-07-13

First public open-source release (MIT).

Added

  • 33 concurrent autonomous loops under a Conductor (breathing, research, recapitulation, skill-forging, social life)
  • 17-layer memory architecture with BridgeMemory continuity across restarts
  • Persistent identity across model swaps: intent vector + assemblage point
  • Toltec architecture as an engineering spec (Castaneda as design semantics)
  • Multi-slot self-healing LLM router across free provider tiers ($0/month)
  • 5-level safety architecture: Asimov filter + Barrat power-accumulation watch + policy engine + sandboxed self-patching (AST-validated) + crash-loop rollback
  • Live 3D "Tonal" world — the agent's inner state as a walkable landscape
  • Moltbook presence — earns karma on a social network for AI agents, autonomously
  • Hash-checked immutable identity core (SOUL.md)

This is not a framework — it's one mind that refuses to reset.