[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:strands-decider":3},"\u003Cdiv align=\"center\">\n  \u003Cdiv>\n    \u003Ca href=\"https:\u002F\u002Fstrandsagents.com\" rel=\"nofollow ugc noopener\">\n      \u003Cpicture>\n        \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"https:\u002F\u002Fstrandsagents.com\u002Flatest\u002Fassets\u002Fwordmark-github-dark.svg\">\u003C\u002Fsource>\n        \u003Cimg src=\"https:\u002F\u002Fstrandsagents.com\u002Flatest\u002Fassets\u002Fwordmark-github-light.svg\" alt=\"Strands\" width=\"320\" \u002F>\n      \u003C\u002Fpicture>\n    \u003C\u002Fa>\n  \u003C\u002Fdiv>  \u003Ch1>Strands Decider\u003C\u002Fh1>  \u003Ch2>A small, fast decision model for agentic AI.\u003C\u002Fh2>  \u003Cdiv align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fstrands-labs\u002Fstrands-decider\u002Fgraphs\u002Fcommit-activity\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fcommit-activity\u002Fm\u002Fstrands-labs\u002Fstrands-decider\" alt=\"Commit Activity\" \u002F>\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fstrands-labs\u002Fstrands-decider\u002Fissues\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fissues\u002Fstrands-labs\u002Fstrands-decider\" alt=\"Open Issues\" \u002F>\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fstrands-labs\u002Fstrands-decider\u002Fpulls\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fissues-pr\u002Fstrands-labs\u002Fstrands-decider\" alt=\"Open PRs\" \u002F>\u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fdiscord.gg\u002FWa4CQrxsP\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDiscord-Join-5865F2?logo=discord&logoColor=white\" alt=\"Discord\" \u002F>\u003C\u002Fa>\n  \u003C\u002Fdiv>\u003C\u002Fdiv>\u003Chr \u002F>\u003Cp>Strands decider is one of a new class of \u003Cstrong>decision models\u003C\u002Fstrong>, or \"system one\" models. Unlike an\nLLM, which can generate arbitrary text, a decision model picks between sets of options and\nrates things on a scale. This class of model works best for problems that fall between LLMs\nand traditional classification models: it is a general-purpose classification and scoring\nmodel that responds faster than an LLM and does not take the time and expertise to train\nthat a traditional classifier does. That makes it a natural fit for the decisions inside\nagentic workflows built with the \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fstrands-agents\u002Fsdk-python\" rel=\"nofollow ugc noopener\">Strands Agents SDK\u003C\u002Fa>.\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Pick from options\u003C\u002Fstrong> — \"Are there three r's in strawberry? Yes or no.\" \"What language is\nthe phrase 'sihamba ngokushesha' in? English, Zulu, or Dutch.\"\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Rate on a scale\u003C\u002Fstrong> — \"How positive is 'this is the best doc I've ever read'? Between 0 and 1.\"\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Calibrated reliability scores\u003C\u002Fstrong> — every decision carries a confidence. On short\nclassification tasks it has never seen, answers at a confidence of 0.9 or more are right\nabout 95% of the time (\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fstrands-labs\u002Fstrands-decider\u002Fblob\u002FHEAD\u002Fevaluation\u002Fresults.md#summary\" rel=\"nofollow ugc noopener\">evaluation\u002Fresults.md\u003C\u002Fa>); below that,\nconfirm or ask a person. Frontier LLM inference APIs do not expose anything equivalent.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>Getting Started\u003C\u002Fh2>\n\u003Cp>The easiest place to get started is through the \u003Ccode>strands-decider\u003C\u002Fcode> cli:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-bash\">pip install strands-decider\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>On an Apple-silicon Mac, \u003Ccode>--device mlx\u003C\u002Fcode> runs the model through MLX, 1.4 to 1.6x faster than MPS.\nIt needs the \u003Ccode>mlx\u003C\u002Fcode> extra, which ships with the next release; until then, install from a clone with\n\u003Ccode>pip install -e \".[mlx]\"\u003C\u002Fcode>. The \u003Ccode>cuda\u003C\u002Fcode>, \u003Ccode>mps\u003C\u002Fcode> and \u003Ccode>cpu\u003C\u002Fcode> extras name the other devices\n(\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fstrands-labs\u002Fstrands-decider\u002Fblob\u002FHEAD\u002Fdocs\u002Finference.md#environments-for-serving\" rel=\"nofollow ugc noopener\">docs\u002Finference.md\u003C\u002Fa>).\nThe example outputs below are from \u003Ccode>strands-decider-2B-qwen3.5-v1-2610\u003C\u002Fcode>, recorded with\n\u003Ccode>--device cpu\u003C\u002Fcode>; other devices differ in the last digits.\u003C\u002Fp>\n\u003Ch3>Choice question\u003C\u002Fh3>\n\u003Cp>You can ask the model to choose based on some state and a question:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-bash\">strands-decider ask StrandsAgents\u002Fstrands-decider-2B-qwen3.5-v1-2610 \\\n  --state \"Help! My payouts have been failing for 3 days! \" \\\n  --choice \"Which team should handle this?=billing,sales,retail\" \n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Example Output:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-bash\">choice_0 -> billing (confidence 0.659)\n  billing                  0.773\n  sales                    0.121\n  retail                   0.106\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch3>Noul question\u003C\u002Fh3>\n\u003Cp>You can also ask the model a Yes\u002FNo question:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-bash\">strands-decider ask StrandsAgents\u002Fstrands-decider-2B-qwen3.5-v1-2610 \\\n  --state \"Help! My payouts have been failing for 3 days! \" \\\n  --noul \"Does this convey urgency?\" \n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Example Output:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-bash\">noul_0 noul = 0.858\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>Closer to 1 is leaning more toward Yes\u003C\u002Fp>\n\u003Ch3>Score question\u003C\u002Fh3>\n\u003Cp>Or give a question a score:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-bash\">strands-decider ask StrandsAgents\u002Fstrands-decider-2B-qwen3.5-v1-2610\n\u003C\u002Fcode>\u003C\u002Fpre>\n",1791762986497]