headroomlabs-ai/headroom
Headroom is a context compression layer for AI agents, reducing token usage by 60-95% across tool outputs, logs, RAG chunks, files, and conversation history, while maintaining accuracy.
Awesome AI Agents › End-to-End Optimizers
algorithmicsuperintelligence/optillm
OptiLLM is an advanced optimizing inference proxy designed to enhance the accuracy and performance of large language models (LLMs) on reasoning tasks without requiring any additional training or fine-tuning. It acts as a drop-in replacement for any OpenAI-compatible API endpoint, enabling users to achieve 2 to 10 times better accuracy on tasks involving math, coding, and logical reasoning by applying over 20 state-of-the-art optimization techniques during inference. These techniques include multi-agent reasoning, chain-of-thought with reflection, planning and optimization strategies, self-consistency methods, and more, which collectively improve the reasoning capabilities of base models significantly. The project supports multiple providers and models, including OpenAI, Anthropic, Google, Cerebras, and over 100 models via LiteLLM, making it versatile and widely applicable. OptiLLM is production-ready and already used by companies and researchers globally. It offers easy installation via pip or Docker, with options for full, proxy-only, or offline usage. The proxy server can be started with minimal configuration, and users can simply redirect their API calls through OptiLLM to benefit from enhanced model performance. OptiLLM demonstrates proven results on various benchmarks, showing substantial improvements in scores on challenging reasoning datasets such as AIME 2025, Math-L5, GPQA-Diamond, InfiniteBench, and others. For example, it can boost the performance of models like Gemini 2.5 Flash Lite, Llama 3.3 70B, and GPT-4o-mini to match or exceed more powerful models by leveraging inference-time compute optimizations. The project also provides SSL configuration options for secure deployment in corporate environments and supports custom CA certificates. Its modular design includes multiple implemented techniques like MARS (Multi-Agent Reasoning System), CePO (Cerebras Planning and Optimization), PlanSearch, ReRead, Z3 Solver integration, and Mixture of Agents, among others, allowing users to tailor the inference process to their specific needs. Overall, OptiLLM is a powerful tool for anyone looking to maximize the reasoning accuracy of LLMs efficiently and effectively without the overhead of retraining models.
https://github.com/algorithmicsuperintelligence/optillm
Also in End-to-End Optimizers
topoteretes/cognee
Cognee is an open-source project that provides scalable and modular memory solutions for AI agents using ECL pipelines, enabling efficient data ingestion, knowledge graph generation, and query capabilities to enhance AI performance and reduce hallucinations.
letta-ai/letta
Letta is an open-source framework for building stateful AI agents with advanced reasoning, transparent long-term memory, and context management, supporting multiple LLM backends and providing a graphical development environment.
langchain-ai/opengpts
OpenGPTs is an open-source platform that replicates and extends OpenAI's GPTs and Assistants API, offering customizable cognitive architectures, extensive tooling, and support for multiple language models and vector databases.
aiwaves-cn/agents
Agents 2.0 is an open-source framework that enables the training and self-evolution of autonomous language agents through symbolic learning inspired by neural network methodologies.
getzep/zep
Zep is an AI memory foundation that uses a temporal Knowledge Graph to enable continuous learning and personalized experiences for AI agents by efficiently managing and retrieving contextual chat and data artifacts.
mll-lab-nu/RAGEN
A reinforcement learning framework for training multi-turn reasoning LLM agents on the StarPO algorithm, shipping ten Gym-compatible environments plus diagnostics for echo-trap and reasoning-collapse failure modes.
langfengQ/verl-agent
verl-agent is an extension of veRL designed for scalable training of large language and vision-language model agents via reinforcement learning, featuring customizable input structures, diverse RL algorithms including GiGPO, and support for multi-modal environments.