extended-thinking

Minimal Go agent framework for LLM tool-calling loops (v1.0, stable). Typed tools, parallel execution, three hooks, streaming progress, snapshot resume, prompt caching across iterations. Anthropic Claude, OpenAI GPT-5, Google Gemini via pi-llm-go.

Recovering verbatim private reasoning from Claude extended-thinking signatures, with a local byte-exact verifier.

Minimal Go LLM client (v1.0, stable). Streaming, strict tool calling, adaptive extended thinking, prompt caching, token counting, cost helpers, retry middleware. Anthropic Claude, OpenAI (GPT-5 + Responses), Google Gemini, OpenAI-compatible hosts.

Reasoning canvas for AI CLIs — captures extended thinking from Claude Code, Codex, Gemini, and Droid as a navigable, forkable tree. Rewind, fork from any node, inject facts mid-thought. Local-first, ships an MCP server.

Active session monitor and infinite-loop breaker for Cursor AI Composer (Agent Mode). Optimizes extended thinking token drain for Claude Opus 4.8 and Fable 5 models during multi-file edits.

Claude extended thinking API patterns: budget management, context compression, batch processing, streaming

Visualize Claude 4.5's internal reasoning!

Single-file Anthropic Claude API playground with streaming, extended thinking, and dark terminal UI

AI-native unified cloud governance platform — multi-cloud discovery, 6R migration planning, IaC security, FinOps, compliance audit, and executive AI chat. Built on Claude Opus 4.7. EU AI Act Annex IV ready. Zero paid SaaS dependencies.

Force maximum reasoning on all Claude Code API calls -- including subagents that run with thinking disabled

A real-time AI transparency tool that sends the same prompt to 2 LLMs simultaneously — visualizing each model's full extended thinking process, web sources selected, and reasoning divergence side by side. Built with Next.js, TypeScript, and the Anthropic API. BYOK + BYOS.

Steelman both sides of any technical decision using Claude Opus 4.7's extended thinking — then decide with confidence