Release 1.0.0rc3 - Pure SQLite & LLM-Only Memory Engine

Release 1.0.0rc3 - Pure SQLite & LLM-Only Memory Engine

Experienced LLMs is a self-learning heuristic ledger and autonomous cognitive consolidation engine for local AI coding agents.

🌟 Key Highlights & Architecture

• Strict 2-File Containment & Zero Workspace Pollution:
  • Contained entirely in ~/.experienced-llms/memory.db and ~/.experienced-llms/EXPERIENCE.md.
  • Zero project repository dotfiles, zero raw daily markdown logs, and zero workspace pollution.
• Pure SQLite Architecture:
  • Thread-safe, high-concurrency WAL mode (PRAGMA journal_mode=WAL).
  • Full-stack migration engine (schema_migrations) managing schema and config evolutions safely.
• Autonomous Interday Consolidation Engine:
  • Pathway A (Tooling / Daemon): Automated nightly consolidation via cron or Windows Task Scheduler.
  • Pathway B (Agentic In-Session Chat Loop): LLMs invoked in chat run experienced-llms consolidate --prepare, reason over active and pending facts in their own context window, and commit via experienced-llms consolidate --apply.
• Zero Token Waste & Token Density:
  • Automated strip filters remove HTML comments and timestamps.
  • Configurable word count ceiling (max_experience_words: 1200, --max-words <int>).
• Customizable Consolidation Algorithm:
  • External prompt template loader supporting global (~/.experienced-llms/consolidation_prompt.txt) and CLI flag (--template <path>) overrides.
• Audited 12-Subcommand CLI Suite:
  • Full suite covering log, consolidate, render, show, status, setup, schedule, prompt, integrate, migrate, init, and --version.
• Standards Compliant:
  • Fully compliant with the agenticskills.io/methodology standard for AI coding agents.

📦 Installation

POSIX (macOS, Ubuntu, Arch Linux, Debian):

curl -fsSL https://github.com/developer88/experienced-llms/releases/download/1.0.0rc3/install.sh | bash

Windows 11 (PowerShell):

irm https://github.com/developer88/experienced-llms/releases/download/1.0.0rc3/install.ps1 | iex

1.0.rc2 - Experienced LLMs

Experienced LLMs 1.0.rc2: Multi-Scheduler Support & Timezone Robustness

Second release candidate of Experienced LLMs — autonomous, self-learning cognitive memory system for AI coding agents.

Changes & Enhancements in RC2

• Configurable Multi-Scheduler Architecture: Added support for selecting between auto, cron, and schtasks via experienced-llms schedule --install --type [auto|cron|schtasks]. Works across native Linux, macOS, WSL, and Windows.
• Windows Task Scheduler & On-Boot Trigger: Added Windows Task Scheduler support bridging WSL execution (wsl.exe -d Ubuntu -- bash -lc ...) and an on-boot background startup script so consolidation passes run even if the PC was turned off during the scheduled 02:00 AM pass.
• Multi-Schedule Status Monitoring: experienced-llms schedule --status dynamically checks and displays whichever schedulers are active on the system (both crontab and Windows Task Scheduler).
• Timezone-Aware Date Filtering: Enhanced SQLite query layer (src/db.py) to support both localtime and UTC strftime('%Y-%m-%d') evaluation around midnight boundaries.
• Repository Cleanup: Removed obsolete prototype demo files from the initial mockup phase.
• Acceptance Verification: Verified end-to-end installation via install.sh, autonomous logging via agy, log inspection, and consolidation passes.

One-Click Installation

# Linux / macOS / WSL
curl -fsSL https://raw.githubusercontent.com/developer88/experienced-llms/main/install.sh | bash

# Windows PowerShell
irm https://raw.githubusercontent.com/developer88/experienced-llms/main/install.ps1 | iex

Full Changelog: 1.0.rc1...1.0.rc2

1.0.rc1 - Experienced LLMs

Experienced LLMs 1.0.rc1: Autonomous Memory & Operational Wisdom for AI Coding Agents

First release candidate of Experienced LLMs — an autonomous, self-learning cognitive memory system for AI coding agents.

Key Capabilities

• Zero-Token Intraday Collector: Deterministic, offline, append-only logger (~/.experienced-llms/raw_days/YYYY-MM-DD.md + SQLite memory.db) recording actionable Reason -> Outcome heuristics.
• Inter-Day Cognitive Consolidator: Autonomous scheduled daemon (02:00 AM) that deduplicates facts, resolves contradictions, and compiles the authoritative ~/.experienced-llms/EXPERIENCE.md.
• Universal Multi-Agent Support: Out-of-the-box integration with Antigravity, Claude Code / Desktop, GitHub Copilot, Cursor, and Pi / generic agents via AGENTS.md.
• One-Click Cross-Platform Installers: Fully automated setups via install.sh (Linux/macOS/WSL) and install.ps1 (Windows with automatic WSL bridging and CLI shim).
• Pluggable LLM Backends: Support for Google Gemini, Anthropic Claude, OpenAI, and local Ollama — complete with a deterministic offline fallback when no API key is configured.
• Battle-Tested Concurrency & Archiving: SQLite WAL mode with busy timeout handling, direct CLI import path resolution, and automatic raw log archiving.

One-Click Installation

# Linux / macOS / WSL
curl -fsSL https://raw.githubusercontent.com/developer88/experienced-llms/main/install.sh | bash

# Windows PowerShell
irm https://raw.githubusercontent.com/developer88/experienced-llms/main/install.ps1 | iex

Full Changelog: https://github.com/developer88/experienced-llms/commits/1.0.rc1