token-saving

OpenCode package for Caveman: terse AI responses, slash commands, compact reviews, commit messages, and markdown memory compression.

ContextCore: An MCP server for Claude (or any AI tool) that enables massive token saving through hybrid search (BM25 + Embeddings)

🦴 paleo — token-saving skills for LLM agents: compress output, trim context, cap budget (Claude Code / Codex / Gemini / Hermes)

Local-first Trello cache with Git-style sync — built and optimised for AI agent workflows.

Project-agnostic dual-memory MCP CLI for Claude Code, Cursor, and OpenCode (Qdrant tuned hybrid retrieval + structural memory hooks)

Comet helps you save 15,000+ tokens! Comet is a TUI application that automatically generates descriptive git commit messages using local language models via Ollama or LMStudio or Openrouter. It analyzes your staged git diffs and provides a clean interface to review, edit, regenerate, and commit your changes instantly.

Auto model-switching plugin for Claude Code — routes prompts to haiku/sonnet/opus (or any custom tier) to save API tokens

High-Trust Open-Core framework for Windows AI Agents. Crushes latency to 3.9ms, enables background UIA invocation, and protects enterprise RPA workflows.

Stop AI agent loops before they burn your tokens. Watchdog for Claude Code / agent CLI transcripts: loop, oscillation, and stall detection with Telegram alerts. Zero dependencies.

Stop hitting Claude Pro's weekly limit by Wednesday. Delegate routine & complex tasks from Claude Opus 4.7 to DeepSeek – slash commands included.

Zero-dependency Q&A cache for OpenClaw - SQLite-based, no Redis/Embedding API needed. Reduce token consumption with keyword matching + edit distance.

AstrBot 对话频率限制插件：限制私聊/群聊在时间窗口内的对话次数，防刷屏与 token 滥用，支持白名单豁免和按用户/群单独设限

WIN10/11 & WSL2 Local File Search Skill | Millisecond Response | Token Saving for AI Agents

AgentFirst - AI/Agent 调 API 的去人化代理内核：响应瘦身(省80%+)、缓存、成本预估、数据飞轮、计费，附带 9 个开箱即用的预配置包 | OpenAPI proxy gateway with response slimming, caching, cost estimation and billing

Auto-trigger graphify knowledge-graph queries on every LLM prompt + MCP shell delegation for Claude Code / Cowork agents.

Lumi (Local Understanding & Model Interface): a beginner-friendly local AI helper for Ollama, Codex and Claude Code

ChangShen - cross-session memory recovery for AI agents: never cold-start again. Hash-indexed conditional reads cut recovery cost ~70% (~55-75% with host-injected memory). Pure local, zero deps.

Local co-processor for AI coding assistants (Claude Code, Aider, Codex). Offload file ops & code tasks to Ollama. Save 60%+ tokens.

Claude Code plugin: Claude routes small coding tasks to the AIs you already have - your Ollama machines and the free tiers of Gemini, Groq, OpenRouter and friends - cheapest first, with quotas, budgets and a self-learning benchmark.