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
TIGER-AI-Lab/verl-tool
VerlTool is a reinforcement learning framework for training tool-using LLM agents, built as an extension of verl and published as a TMLR 2026 paper that also received a best paper award at the ICLR 2026 SPOT workshop. Its central design choice is complete decoupling of actor rollout from environment interaction: verl is kept as a git submodule so upstream improvements flow in, while all tool calls go through one unified API. Adding a new tool means adding a single Python file that can be tested on its own, without touching the training loop. Tools are treated as environments under a tool-as-environment paradigm, meaning each interaction can mutate environment state, and that state is stored and reloaded per trajectory so multi-turn episodes stay consistent. The framework natively supports interactive loops between an agent and its tool environment rather than single-shot generation, and it implements trajectory-level asynchronous rollout, which the authors report speeds up generation with tool calling by at least a factor of two. Tools run behind a separate tool server that can be exercised independently of training. The repository ships training recipes for several agentic tasks, including a reproduction of Search-R1 that the maintainers report exceeds the original on the same benchmarks, natural-language-to-SQL tool training, and the ToRL and DAPO recipes. The codebase was reorganized for modularity and tracks recent verl and vLLM releases, with upgrade notes documenting the migration. Installation instructions, training recipes and design documentation live in the repository, supplemented by generated DeepWiki docs, a Discord server and a WeChat group. It is aimed at researchers and engineers doing reinforcement learning post-training of agents that must call external tools, rather than at application developers assembling agents from pretrained models.
https://github.com/TIGER-AI-Lab/verl-tool
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.
algorithmicsuperintelligence/optillm
OptiLLM is an optimizing inference proxy that significantly improves the accuracy and performance of large language models on reasoning tasks without requiring any model training, by applying over 20 advanced optimization techniques during inference.
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.