Skip to content

Releases: Alcheri/Geminoria

v1.1.0-beta.4

Choose a tag to compare

@Alcheri Alcheri released this 19 Apr 06:42

🌐 2026-04-19 - Network-Aware History Tool Allowlists (v1.1.0-beta.4)

  • Bumped release version from 1.1.0-beta.3 to 1.1.0-beta.4.
  • Changed historyToolsChannelAllowlist to a network-scoped setting.
  • Changed searchLastChannelAllowlist to a network-scoped setting.
  • Changed searchUrlsChannelAllowlist to a network-scoped setting.
  • Updated runtime tool gating to read those allowlists from the active IRC network context.
  • Updated README examples to show explicit per-network config paths.

⚠️ Beta notes

Choose a tag to compare

@Alcheri Alcheri released this 13 Apr 08:13
  • Geminoria is now published as a SemVer beta release (1.1.0-beta.3).
  • Internal architecture now supports modular service/core/memory/cache separation.
  • Please report any issues with tool-calling behaviour, config detection, or Gemini responses.

🏷️ 2026-04-13 - SemVer Release Tag (v1.1.0-beta.3)

  • Bumped release version from 1.1.0-beta.2 to 1.1.0-beta.3.
  • Updated package meta-data and docs version tags to 1.1.0-beta.3.
  • Finalised architecture split baseline for future collaborative modules (core.py, services.py, memory.py, cache.py).
  • Re-factor plugin into package layout and finalise phase-2 clean-up.

Geminoria v1.1.0-beta.1

Pre-release

Choose a tag to compare

@Alcheri Alcheri released this 31 Mar 11:35

Agentic Gemini‑powered search for Limnoria — Beta Release

This beta introduces a major architectural upgrade to Geminoria, shifting from environment‑based configuration to a fully Limnoria‑native, config‑driven, agentic design. It is feature‑complete for the 1.1 series and ready for wider testing.

✨ What’s new

Config‑driven Gemini integration

  • Gemini API key now stored in Limnoria config: supybot.plugins.Geminoria.apiKey
  • Model selection also moved to config: supybot.plugins.Geminoria.model
  • Hot‑reload behaviour: client automatically refreshes when the key changes.

Improved agentic loop

  • More robust round‑based tool execution.
  • Cleaner extraction of function calls and text responses.
  • Final synthesis step added when tool‑call limit is reached.
  • Graceful fallback to tool results if Gemini returns no text.

Enhanced config search

  • New config walker returns (full_path, is_leaf) pairs.
  • Leaf keys are prioritised for relevance.
  • Output formatted as a clean, readable list for IRC.
  • Automatic highlighting of exact config keys in responses.

Better logging

  • Round‑by‑round agentic debugging.
  • Tool call tracing with summarised output.
  • Capability‑denial logging.
  • Client refresh logging.
  • Compact text summariser for log readability.

Cleaner SDK integration

  • Added _schema(),_tool(), and _gen_config() helpers to avoid Pylance issues with generated models.
  • System instruction rewritten for clarity and IRC‑friendly output.

🛠️ Fixes & refinements

  • More predictable whitespace and markdown cleaning.
  • URL and message buffers remain lightweight and bounded.
  • Capability checks now log denials for easier debugging.
  • Safer handling of missing candidates or malformed responses.
  • Improved fallback behaviour when Gemini returns empty content.

⚠️ Beta notes

  • This is a pre‑release intended for real‑world testing.
  • Behaviour may change slightly before the final 1.1.0 release.
  • Please report any issues with tool‑calling behaviour, config detection, or Gemini responses.