v0.1.3

Changelog

All notable changes to this project will be documented in this file.

[0.1.3] — 2025-04-02

Fixed

• README now displays correctly on PyPI dashboard (re-release of 0.1.2)

[0.1.2] — 2025-04-02

Fixed

• README now displays correctly on PyPI dashboard

[0.1.1] — 2025-04-02

Changed

Dependency optimization

• Made sentence-transformers optional — base install now only requires numpy (~10MB vs ~1.5GB)
• Added pip install clm-plugin[embed] for full embedding support
• Auto-detect missing sentence-transformers and enable no_embed=True with helpful message

PyPI metadata improvements

• Added authors and maintainers fields
• Added Repository and Changelog URLs
• Created PyPI-specific README (README_PYPI.md) — short, install-focused, under 100 lines
• Full README.md remains on GitHub

Documentation

• Added note about sidecar database (*.db) exclusion from version control
• Clarified that *.db files should not be committed

Fixed

• Prevented crashes for users who install lean version without reading docs

[0.1.0] — 2025-03-31

First public release

Core architecture

• 5-layer cognitive load management: Signal Collector, CLM Scorer, Chunking Engine, Action Dispatcher, Sidecar Store
• 4 cognitive load signals: branching factor, repetition rate, uncertainty density, goal distance
• 3-zone intervention system: Green (pass), Amber (compress), Red (interrupt)
• Abort action for structurally unresolvable tasks (5 consecutive Red triggers)
• Amber escalation protection (3 consecutive Amber triggers → Red)

Integration

• CLM() — zero-argument instantiation with sensible defaults
• observe_raw() — single-line integration, no TaskState construction required
• AutoStateBuilder — automatic task tree inference from LLM outputs
• LangChain adapter: CLMCallbackHandler
• OpenAI Agents SDK adapter: CLMOpenAIHook
• Generic loop adapter: CLMLoop with decorator and context manager support

Observability

• verbose=True — real-time step-by-step output
• get_history() — full intervention log
• summary() — session aggregate stats
• get_score(), get_zone(), get_sidecar_stats()

Configuration

• no_embed=True — keyword-based fallback, zero model download, works offline
• Fully tunable weights, thresholds, and zone boundaries
• Domain-specific configuration examples: medical, legal, voice

Storage

• SQLite sidecar store, auto-created on first use
• In-memory mode (default) for ephemeral sessions

Known limitations

• Default weights [0.30, 0.25, 0.25, 0.20] are heuristic, not empirically validated
• AutoStateBuilder uses regex heuristics for task tree inference
• response.context replaces task plan section only, not full conversation history
• Embedding model requires ~90MB download on first use (avoidable with no_embed=True)

CLM v0.1.0 First Public Release

CLM v0.1.0 — First Public Release

Real-time metacognitive middleware for LLM agents. One line to add to any agent loop.

Install

pip install clm-agent

Quickstart

from clm import CLM
clm = CLM(verbose=True)
result = clm.observe_raw(llm_output)

What's in this release

• 4 cognitive load signals (branching, repetition, uncertainty, goal drift)
• 3-zone intervention system with Amber escalation and abort safety
• LangChain, OpenAI Agents, and generic loop adapters
• no_embed mode for offline environments
• Domain configs for medical, legal, and voice agents

Known limitations in v0.1

• Default weights are heuristic — tune them for your domain
• No async support yet (v0.2)
• No CrewAI/AutoGen adapters yet (v0.2)

Full docs in README. Issues and usage reports welcome.