A developer introduced DecisionLoop, a Python framework where a small decision model handles intent detection, tool selection, arguments, and control flow rather than calling an LLM for every decision. Early tests on three use cases showed better tool-call accuracy and lower cost and latency than a LangChain LLM agent using a 12B model. The results are preliminary, and the author is seeking feedback on architecture and benchmarking.
A developer released DecisionAct, an open-source Python framework where predefined agents use decision models to interpret natural-language requests, call tools, and select allowed outcomes rather than relying on an LLM to generate every response. Python functions execute actions and predefined templates produce final replies, positioning the framework as a faster, cheaper option for workflows with predefined actions and outcomes. It ships via PyPI with examples and is soliciting community feedback on its early release.
A developer released DecisionAct, an open-source Python framework where predefined agents use decision models to interpret natural-language requests, call tools, and select allowed outcomes rather than relying on an LLM to generate every response. Python functions execute actions and predefined templates produce final replies, positioning the framework as a faster, cheaper option for workflows with predefined actions and outcomes. It ships via PyPI with examples and is soliciting community feedback on its early release.
The developer of DecisionAct, an open-source Python framework for task-focused agents driven by decision models (JEV, Cloudflare Clef) rather than LLMs, announced v0.1.2. The update improves multi-tool call reliability: repeated tool calls resolve fresh inputs, earlier tool results can feed later calls, and unresolved input selections trigger a user query. Practitioners can install it via pip and follow quickstart examples in the docs.
The developer of DecisionAct, an open-source Python framework for task-focused agents driven by decision models (JEV, Cloudflare Clef) rather than LLMs, announced v0.1.2. The update improves multi-tool call reliability: repeated tool calls resolve fresh inputs, earlier tool results can feed later calls, and unresolved input selections trigger a user query. Practitioners can install it via pip and follow quickstart examples in the docs.
Z-Engine is an open source coding agent that uses LLMs for understanding problems, investigating repos, and writing plans. The builder is experimenting with Laya, a smaller model, to handle frequent small decisions around the agent, like tool-call approval, task routing, escalation, and context relevance. It targets developers building coding agents who want to offload micro-decisions from expensive LLM calls.
Z-Engine is an open source coding agent that uses LLMs for understanding problems, investigating repos, and writing plans. The builder is experimenting with Laya, a smaller model, to handle frequent small decisions around the agent, like tool-call approval, task routing, escalation, and context relevance. It targets developers building coding agents who want to offload micro-decisions from expensive LLM calls.
A lightweight router that uses TypeSafe's Jev model to decide tool calls and agent paths quickly. Jev returns typed Choice and Score primitives instead of generated text, cutting latency and cost versus generative LLM routing. Aimed at developers building low-latency agent architectures.
A lightweight router that uses TypeSafe's Jev model to decide tool calls and agent paths quickly. Jev returns typed Choice and Score primitives instead of generated text, cutting latency and cost versus generative LLM routing. Aimed at developers building low-latency agent architectures.
A developer released harness-router, an open-source MIT-licensed MCP server that routes tool calls for agentic coding harnesses. It keeps obvious actions cheap, uses a small decision model (Jev) for ambiguous choices, and runs Monte Carlo Tree Search when multi-step consequences matter. The author reports exploratory benchmark results against Codex tool selection, though the MCTS budget was tuned on the same test cases.
A developer released harness-router, an open-source MIT-licensed MCP server that routes tool calls for agentic coding harnesses. It keeps obvious actions cheap, uses a small decision model (Jev) for ambiguous choices, and runs Monte Carlo Tree Search when multi-step consequences matter. The author reports exploratory benchmark results against Codex tool selection, though the MCTS budget was tuned on the same test cases.