Five local open-source repos that can replace Jev-style decision calls
A roundup of local repos including rizzo-flow and llm2jev that return structured yes/no or scored decisions instead of Jev-style calls, with a cited 48ms versus 652ms latency comparison.
this is pure f*cking treasure. 5 GitHub repos worth checking before you add another AI subscription. not another chatbot wrapper. these target the routine work you keep sending to expensive models: 1. rizzo-flow a local alternative for Jev-style decisions. yes / no pick an option return a score the default model is around 4.4GB and runs through llama.cpp. it returns structured decisions with probabilities instead of generating a paragraph. a compatible HTTP interface lets you redirect supported clients to localhost by changing the endpoint. 2. llm2jev turns Qwen3.5-4B into a structured decision model using a single forward pass. no token-by-token answer generation. the maintainers report 48ms median latency in their benchmark, compared with 652ms for the Jev version they tested. that’s a specific benchmark, not a promise for every machine. still worth testing for high-volume routing. 3. fast-browser-use local browser automation built around models like Qwen3.5-9B. the interesting part: routine browser actions can use local inference instead of calling a paid cloud model at every step. useful when your agent spends more time clicking through pages than doing difficult reasoning. 4. DeepSeekGUI a desktop workbench with local workspaces, Git, a built-in browser and memory. connect through a DeepSeek account or your own model API key. less tab juggling. more of the coding workflow in one place. 5. oriveo one chat client for 15 providers, including OpenAI, Anthropic, Gemini, Grok and DeepSeek. also supports local endpoints like Ollama and LM Studio. your keys. your choice of model. conversations stored locally. paid API usage still costs money. local models still need hardware. none of this guarantees frontier-model quality. but paying frontier-model prices to classify text, choose an option or handle routine clicks deserves a rethink. keep the expensive model for the hard problem. give the boring work a cheaper route.
This post quotes another post: x.com/ggwplabs/status/2106004888312217829