#python #ai_coding #claude #claude_code #code_review #graphrag #incremental #knowledge_graph #llm #mcp #python #static_analysis #tree_sitter

code-review-graph is a free, local-first tool that builds a structural map of your code to stop AI assistants from re-reading your entire codebase, saving you significant money and time. By using Tree-sitter to parse your code into a graph and tracking changes incrementally, it identifies the exact "blast radius" of a change so your AI only reads the few files that actually matter. This reduces token usage by an average of 82x (up to 528x in best cases), cutting costs and making AI reviews feasible for massive projects like monorepos.

https://github.com/tirth8205/code-review-graph

#python #agent #agent_memory #ai_agent #book #coding_agent #context_engineering #large_language_models #llm #mcp #multi_agent #multimodal #rag #reinforcement_learning

This free open-source book and code library teaches you how to build AI Agents using the simple formula: Agent = Model + Context + Tools. You get the full book in English, Chinese, and Tamil, plus ready-to-run code examples for every chapter. The benefit is that you can learn by doing instead of just reading, letting you build real AI agents that search the web, write code, remember users, and work with other agents to solve complex tasks.

https://github.com/bojieli/ai-agent-book

#typescript #a2a #ai_agents #ai_gateway #anthropic #claude #claude_code #cline #codex #copilot #cursor #deepseek #free_ai #gemini #kimi #llm_gateway #mcp #openai #openai_proxy #qwen #token_saver

OmniRoute is a free tool that connects your AI coding apps to 268 providers, including 90+ with free tiers, so you can use top models like Claude or GPT without paying. It automatically switches to another provider if one hits its limit and compresses your requests to cut token usage by up to 95%. You benefit by saving money, avoiding interruptions, and getting unlimited access to powerful AI tools through a single, easy setup.

https://github.com/diegosouzapw/OmniRoute

#rust #gguf #llm #localai #mlx #skill #unsloth

llmfit is a free terminal tool that automatically checks your computer’s RAM, CPU, and GPU to tell you exactly which large language models will run well on your machine, ranking them by fit, speed, quality, and context. It supports both an interactive visual interface and a command-line mode, works with multi-GPU setups, and suggests the best quantization levels so you never waste time or storage downloading models that won’t work. This saves you from guesswork, trial-and-error testing, and wasted bandwidth, giving you immediate, reliable recommendations for the best local AI models for your specific hardware.

https://github.com/AlexsJones/llmfit

#go #agent #agent_framework #ai_agent #ai_coding #cli #coding_agent #deepseek #developer_tools #ink #llm #prompt_caching #r1 #terminal #tool_use #tui #typescript

Reasonix is a DeepSeek-based AI coding agent for your terminal, desktop, or VS Code, and it uses one local engine for all three. You can set it up with a simple config, add plugins and tools, and run it as a single Go binary, which helps you code faster while keeping token costs lower during long sessions.

https://github.com/esengine/DeepSeek-Reasonix

#python #agent_security #ai_agents #ai_security #benchmark #claude #claude_code #codex #cursor #llm_security #mcp #model_context_protocol #prompt_injection #threat_detection

ADR is an enterprise security system for AI agents that watches what agents do, tests them for attacks, detects risky behavior, and is being used in production at Uber; its paper was accepted to MLSys 2026. The benefit to you is better protection for AI tools like Cursor, Claude Code, and Codex, plus a way to find unsafe actions before they cause harm.

https://github.com/uber/ADR

#python #agent_memory #ai #ai_governance #ai_infrastructure #artificial_intelligence #context_engineering #context_graphs #data_engineering #decision_intelligence #developer_tools #explainable_ai #generative_ai #graph_rag #knowledge_graph #llm #ontology #provenance #python #reasoning #semantic_search

Semantica is an open-source tool that turns messy data into clear, connected knowledge graphs for AI. It helps you track where facts came from, explain AI decisions, find conflicts, and keep systems auditable and trustworthy. This benefits you by making AI easier to trust, easier to check, and safer to use in important work like finance, healthcare, legal, and government.

https://github.com/semantica-agi/semantica

#python #ai #ast #claude_code #code_analysis #code_understanding #codebase_search #developer_tools #graph_database #knowledge_graph #llm #mcp #mcp_server #memgraph #monorepo #multi_language #python #rag #retrieval_augmented_generation #semantic_search #tree_sitter

Code-Graph-RAG is an open-source tool that reads a codebase, builds a graph of functions, classes, modules, and links, and lets you ask questions, find code, edit safely, and improve code in plain English. It supports many languages and can run as an MCP server for direct use with AI tools. Benefit: you can understand and change large mixed-language projects faster, with less manual searching and fewer mistakes.

https://github.com/vitali87/code-graph-rag

#python #cactus #gemini #gemma #llm #on_device_ai

Needle 2 is a very small tool-calling AI model that can run on low memory, work offline, and return structured results like JSON instead of free text. It helps you build apps that call tools, extract data, and fine-tune models more safely and easily, even on devices with limited RAM.

https://github.com/cactus-compute/needle

#typescript #agent #agent_harness #agent_os #agentic #ai #ai_agent #ai_agents #artificial_intelligence #claude_code #codex #electron #holaboss #holaos #llm #mcp #memory #model_context_protocol #runtime #typescript #workspace

holaOS is a local-first desktop workspace that lets you run agents like Claude Code, Codex, or its built-in agent together with your tools, files, apps, and shared memory. It supports macOS, Windows, and Linux, includes built-in models or your own keys, and can help you get work done with real browser control, app automation, and ready-made files. This can save you time because one setup works across agents, keeps your context, and lets you produce useful results without switching systems.

https://github.com/holaboss-ai/holaOS

#python #cli #consumer_gpu #dpo #fine_tuning #gguf #huggingface #llm #llmops #local_ai #local_llm #lora #low_vram #machine_learning #ollama #peft #python #pytorch #qlora #sft #transformers

Soup is a tool that helps you fine-tune and post-train large language models with one command, using one config file and little setup. It can work on a local GPU, even a 4 GB laptop GPU for an 8B model with layer streaming, so you can train without SSH or cloud hassle. This saves time, reduces setup pain, and makes model training easier to start and manage.

https://github.com/MakazhanAlpamys/Soup

#python #agent #agent_security #ai_infra #ai_red_teaming #ai_security #llm #llm_evaluation #llm_jailbreak #llm_security #mcp_scan #openclaw_security #prompt_injection #prompt_security #scanner #security #security_tools #skill_scanner #skills_security #vulnerability

A.I.G (AI-Infra-Guard) is a Tencent Zhuque Lab tool for AI red teaming and security checks. It scans AI infrastructure, MCP servers, agent skills, and jailbreak risks, and it can run with Docker, a web UI, or an API. This helps you find weak points early, understand real security risks, and fix problems before they affect your AI systems.

https://github.com/Tencent/AI-Infra-Guard

#typescript #agent_runtime #ai #ai_agent #apache #cli #desktop #electron #event_sourcing #llm #local_first #maka #tool_use #typescript

Maka is a local-first agent workspace that runs on your computer, lets you use your own model connection, and can read files, run tools, and save recoverable work records. This helps you keep data local, control permissions, and recover or continue work more safely.

https://github.com/apache/maka