SignalTech/AI
DeepSeek v4.1 flash on OpenCode pretty good
DeepSeek v4.1 flash praised on OpenCode, signaling positive AI model reception.
Context
What happened: A Show HN post describes shev, a command line tool that takes one pasted message and returns a single JSON decision with fields actionable, take_action, and confidence. It runs on DeepSeek over the network or a local MLX model on Apple Silicon.
Key numbers: confidence is an integer always in the range 34 to 100.
Why it matters: The author positions it against agent frameworks and prompt engineering, offering one message in and one decision out, with decisions logged to responses.json and readable by other models over MCP.
What changed
A Show HN post describes shev, a command line tool that takes one pasted message and returns a single JSON decision with fields actionable, take_action, and confidence. It runs on DeepSeek over the network or a local MLX model on Apple Silicon.
Why it matters
The author positions it against agent frameworks and prompt engineering, offering one message in and one decision out, with decisions logged to responses.json and readable by other models over MCP.