Paritok-4B-v1

Paritok is an open-source compression gateway for AI coding agents. It runs as a drop-in proxy between an agent and an upstream LLM API, stripping irrelevant tool schemas, compressing tool results and file reads, and summarizing stale conversation history before forwarding requests. Compressed content remains recoverable on demand through the gateway, while responses pass back to the agent unchanged. It supports agents such as Claude Code, Cursor, Codex, and OpenHands, as well as agents that support BASE_URL configuration, and is powered by an open-source code-focused 4B compression model.

Needle 2 is an open-source, 45-million-parameter foundation model for tool calling, device use, and structured extraction on small devices. Its weights are compressed into a single 14 MB binary and integrated with its inference engine; the repository reports that a full session uses about 28 MB of RAM and performs inference without network access after setup. The model converts text and tool descriptions into schema-constrained JSON calls using a byte-level grammar, attaches a learned confidence score for threshold-based escalation, and can retrieve the top five tools from a larger catalogue. It uses a 256-token sliding window with tools pinned as key-value sinks to bound memory, and its Simple Attention Network combines a Hadamard MLP, grouped-query attention, engram key-value memory, and multi-lane hyper-connections. The Python package supports inference, LoRA fine-tuning, and checkpoint export; the inference engine is fetched from Hugging Face and cached, while the training stack is an optional installation.

OKF Agent Memory is a Git-native persistent memory layer for AI coding agents, developed as a pure Go CLI and library. It stores project knowledge as plain Markdown files with YAML front matter in a repository, following the vendor-neutral Open Knowledge Format (OKF) v0.2 rather than relying on an external vector database. The tooling parses and validates OKF bundles, builds an in-memory BM25 index for local lexical search, and provides progressive disclosure through hierarchical index files and link graphs. Its search-before-write workflow queries existing concepts before creating new ones; concepts can include provenance, trust tiers, lifecycle metadata, and code references for linking knowledge to source files. The standalone executable also supports creating and updating concepts, bootstrapping the memory structure into a project, and running an embedded Model Context Protocol (MCP) server over stdio for agent platforms. The repository reports sub-300-microsecond concept search and approximately 4-millisecond graph validation for its benchmark cases, with no external databases or API costs for retrieval. It is distributed under the MIT License.

Terminal Bench 4 is a benchmark for comparing coding-model performance across tasks with different difficulty levels.

An Anthropic reference release containing 36 drop-in inference-optimization kits for open protein- and genomics-machine-learning tools, covering structure prediction, cofolding, binder and sequence design, protein-language-model inference, and genomic prediction. Each kit carries a pinned upstream release in a stock directory and applies optimizations under a named mode without changing the upstream calling interface. Modes include off for the unmodified upstream tool, exact for identical outputs with faster inference, fast for documented numerical differences, and big for lower peak GPU memory; some kits also support splitting a prediction across multiple GPUs. The kits share runtime code for mode resolution, determinism, memory handling, multi-GPU execution, and GPU kernels, and can be installed through Docker, Apptainer, or a pinned Python environment. The repository is a non-maintained reference release accompanying Anthropic's biomolecular-modeling blog post; its original code is under Apache License 2.0, while the carried upstream projects and model weights have their own licenses and terms.