{"slug":"needle-3","name":"Needle 3","summary":"Needle 3 is Cactus Compute’s on-device “automation foundation model”: a laddered / sliceable tiny model (post: 8–29 MB CQ2-bit binary; 25–121M params; depths 2–20 layers from one weight set) built on their Simple Attention Networks, aimed at tool calls, structured extraction, and local text embeddings — not open-ended chat.","kind":"model","homepage":"https://cactuscompute.com/needle","repoUrl":"https://github.com/cactus-compute/needle","publisher":"","hasMcp":false,"hasApi":false,"costRange":"unknown","composabilityRationale":"","functions":["embed","extract"],"functionProbabilities":[{"functionId":"embed","value":0.72,"source":"06 Knowledge Library/Resources/coding/2026-09-18-cactus-compute-needle-3.md","method":"lexical","observedAt":"2026-09-18T00:00:00.000Z"},{"functionId":"extract","value":0.72,"source":"06 Knowledge Library/Resources/coding/2026-09-18-cactus-compute-needle-3.md","method":"lexical","observedAt":"2026-09-18T00:00:00.000Z"}],"slices":[{"functionId":"embed","label":"Embed","note":"Needle 3 is Cactus Compute’s on-device “automation foundation model”: a laddered / sliceable tiny model (post: 8–29 MB CQ2-bit binary; 25–121M params; depths 2–20 layers from one weight set) built on their Simple Attention Networks, aimed at tool calls, structured extraction, and local text embeddings — not open-ended chat.","cost":"unknown"},{"functionId":"extract","label":"Extract","note":"Needle 3 is Cactus Compute’s on-device “automation foundation model”: a laddered / sliceable tiny model (post: 8–29 MB CQ2-bit binary; 25–121M params; depths 2–20 layers from one weight set) built on their Simple Attention Networks, aimed at tool calls, structured extraction, and local text embeddings — not open-ended chat.","cost":"unknown"}],"tweets":[],"sources":[{"kind":"human","label":"Compendium resource card","excerpt":"06 Knowledge Library/Resources/coding/2026-09-18-cactus-compute-needle-3.md\nNeedle 3 is Cactus Compute’s on-device “automation foundation model”: a laddered / sliceable tiny model (post: 8–29 MB CQ2-bit binary; 25–121M params; depths 2–20 layers from one weight set) built on their Simple Attention Networks, aimed at tool calls, structured extraction, and local text embeddings — not open-ended chat.","at":"2026-09-18T00:00:00.000Z"}],"interfaces":[],"origin":"seed","notes":[],"createdAt":"2026-09-18T00:00:00.000Z","updatedAt":"2026-09-18T00:00:00.000Z","stagingSeed":"leyline-0.3.0-staging","recordId":"cmp_needle-3","claims":{"name":"Needle 3","summary":"Needle 3 is Cactus Compute’s on-device “automation foundation model”: a laddered / sliceable tiny model (post: 8–29 MB CQ2-bit binary; 25–121M params; depths 2–20 layers from one weight set) built on their Simple Attention Networks, aimed at tool calls, structured extraction, and local text embeddings — not open-ended chat.","kind":"model","publisher":"","homepage":"https://cactuscompute.com/needle","repoUrl":"https://github.com/cactus-compute/needle","costRange":"unknown","functions":["embed","extract"],"slices":[{"functionId":"embed","label":"Embed","note":"Needle 3 is Cactus Compute’s on-device “automation foundation model”: a laddered / sliceable tiny model (post: 8–29 MB CQ2-bit binary; 25–121M params; depths 2–20 layers from one weight set) built on their Simple Attention Networks, aimed at tool calls, structured extraction, and local text embeddings — not open-ended chat.","cost":"unknown"},{"functionId":"extract","label":"Extract","note":"Needle 3 is Cactus Compute’s on-device “automation foundation model”: a laddered / sliceable tiny model (post: 8–29 MB CQ2-bit binary; 25–121M params; depths 2–20 layers from one weight set) built on their Simple Attention Networks, aimed at tool calls, structured extraction, and local text embeddings — not open-ended chat.","cost":"unknown"}],"interfaces":[],"composabilityRationale":""},"relations":[],"signals":[],"claimVerification":[],"explain":{"score":3,"formula":"100 × sum(weight × value) over observed factors only. Unobserved factors are omitted, not zeroed.","factors":[{"id":"github_health","label":"GitHub health","weight":0.18,"why":"insufficient evidence","sources":[],"evidence":"insufficient"},{"id":"agent_use","label":"Agent use","weight":0.16,"why":"insufficient evidence","sources":[],"evidence":"insufficient"},{"id":"human_notes","label":"Human notes","weight":0.14,"why":"insufficient evidence","sources":[{"kind":"human","label":"Compendium resource card","excerpt":"06 Knowledge Library/Resources/coding/2026-09-18-cactus-compute-needle-3.md\nNeedle 3 is Cactus Compute’s on-device “automation foundation model”: a laddered / sliceable tiny model (post: 8–29 MB CQ2-bit binary; 25–121M params; depths 2–20 layers from one weight set) built on their Simple Attention Networks, aimed at tool calls, structured extraction, and local text embeddings — not open-ended chat.","at":"2026-09-18T00:00:00.000Z"}],"evidence":"insufficient"},{"id":"composability","label":"Composability","weight":0.16,"why":"insufficient evidence","sources":[],"evidence":"insufficient"},{"id":"checklist","label":"Checklist","weight":0.1,"value":0,"contribution":0,"why":"No MCP surface is declared. No API is declared. Cost is unknown. 0 of 3 checklist fields are complete.","sources":[],"evidence":"observed"},{"id":"evidence_quality","label":"Evidence quality","weight":0.14,"value":0.18107144632819594,"contribution":0.025350002485947435,"why":"1 sources cover 1 evidence types. The latest dated evidence is 2026-09-18. Freshness is omitted unless observed as a signal.","sources":[{"kind":"human","label":"Compendium resource card","excerpt":"06 Knowledge Library/Resources/coding/2026-09-18-cactus-compute-needle-3.md\nNeedle 3 is Cactus Compute’s on-device “automation foundation model”: a laddered / sliceable tiny model (post: 8–29 MB CQ2-bit binary; 25–121M params; depths 2–20 layers from one weight set) built on their Simple Attention Networks, aimed at tool calls, structured extraction, and local text embeddings — not open-ended chat.","at":"2026-09-18T00:00:00.000Z"}],"evidence":"observed"},{"id":"benchmark","label":"Benchmarks","weight":0.12,"why":"insufficient evidence","sources":[],"evidence":"insufficient"}],"missing":["GitHub health: insufficient evidence","Agent use: insufficient evidence","Human notes: insufficient evidence","Composability: insufficient evidence","Benchmarks: insufficient evidence"]},"score":3}