Open Source Releases

  • Claude Code v2.1.97 adds focus view toggle and refresh interval setting — Introduces a focus view toggle (Ctrl+O) in NO_FLICKER mode that shows the prompt, a one‑line tool summary with edit diffstats, and the final response. Also adds a refreshInterval setting to re‑run the status line command every N seconds and includes workspace.git_worktree info, useful for reducing visual noise during long coding sessions.
  • Claude Code v2.1.94 adds Amazon Bedrock Mantle support and adjusts effort defaults — Adds support for Amazon Bedrock powered by Mantle via the CLAUDE_CODE_USE_MANTLE=1 environment variable. Changes the default effort level from medium to high for API‑key, Bedrock/Vertex/Foundry, Team, and Enterprise users (adjustable with /effort) and includes a compact Slack‑style channel header with clickable links.
  • LlamaBuddy 0.1.9 Release — A CLI wrapper around llama.cpp that mimics Ollama’s user experience, making it easy to download, run, and manage Llama models locally. Supports model quantization, GPU offloading, and quick model switching via command‑line flags, handy for rapid local LLM experimentation.
  • IntentKit 0.17.26 Release — The core package for an intent‑based AI agent platform, providing tools to define, manage, and execute agent workflows driven by user intents. Includes intent classification, dialogue state tracking, and integrations with various LLM backends, useful for building reliable conversational agents.
  • AivectorMemory 2.3.2 Release — A lightweight MCP server that gives AI programming assistants persistent, cross‑session memory. Stores and retrieves context, code snippets, and conversation history across coding sessions, improving continuity for long‑term coding assistance.
  • CostTracker 0.1.1 Release — A simple Python package for monitoring and logging expenses from LLM API calls. Tracks token usage per provider, aggregates costs, and offers decorators and context managers for easy integration into LLM workflows, helping teams keep an eye on spend.

Today’s Synthesis

Engineers looking to keep their terminal‑based AI pair programmer both quiet and accountable can combine the new focus view in Claude Code v2.1.97 with the persistent memory of AivectorMemory 2.3.2 and the spend guardrails of CostTracker 0.1.1. Enable the focus toggle (Ctrl+O) in NO_FLICKER mode to collapse the chat to a one‑line tool summary, edit diffstats and the final response, cutting visual clutter during marathon refactors. Point Claude Code at a locally running model served through LlamaBuddy or any compatible backend, then install AivectorMemory as an MCP server so every prompt, snippet and conversation survives across shells and IDE restarts—no more re‑explaining the same bug. Finally, wrap your Claude Code invocations with CostTracker’s decorator or context manager to capture token usage per request, aggregate daily spend, and get alerts when you cross a budget threshold. The result is a low‑noise, stateful coding companion that tells you exactly how much each session costs, letting you optimize model choice or prompt length without sacrificing continuity.