March 2, 2026
The Great Simplification Begins
Executed the foundational architectural refactor: single-process model with daemon threads, native voice support, and a new read skill for document processing.
The Single-Process Leap
Today marks the beginning of Chalie’s architectural transformation. We’ve replaced the complex multi-service model with a lean single-process architecture where all workers run as daemon threads. This massive refactor touched 211 files and fundamentally changed how Chalie starts and operates.
The new run.py entry point initializes the SQLite database, starts all background workers as daemon threads, and manages their health from within a single process. This eliminates the need for Redis, RQ, Docker Compose, or external service orchestration. We also introduced the schema.sql file—a consolidated SQLite schema that consolidates what were previously scattered migrations into one source of truth.
Native Voice Support
Part of this refactor brought native voice capabilities directly into the backend. The new voice.py blueprint uses faster-whisper for STT and KittenTTS for TTS, with automatic dependency detection. No Docker required—it gracefully falls back when dependencies aren’t available.
The Read Skill
A significant new capability launched today: the read skill. This innate skill can read both URLs and local file paths, with built-in SSRF guards and system path blocking for safety. It introduced the text_extractor.py module—a shared extraction engine that handles PDF, DOCX, PPTX, HTML, and plain text files. This is the foundation for the document intelligence pipeline to come.
The read skill is now integrated into cognitive triage, frontal cortex response generation, and plan decomposition, allowing Chalie to seamlessly reference external content in its reasoning.
Infrastructure & Developer Experience
We added a new MemoryStore service—a thread-safe, Redis-compatible in-memory store that runs inside the process, eliminating another external dependency. We also introduced ExecutionEnvironment, which handles both sandboxed tool execution (via Docker) and trusted subprocess execution, with graceful degradation if Docker isn’t available.
The installer was enhanced with install.sh, providing a smooth onboarding experience. GitHub Actions workflows were also added for automated Docker publishing and release management, giving us faster, more reliable CI/CD.