Skip to content

Releases

local-jev v1.0.0: Flagship Production 1.0 Release

Choose a tag to compare

@ronitgupta138 ronitgupta138 released this 10 Oct 17:35
53cd292

🚀 local-jev v1.0.0 Production Flagship Release

Zero API keys. Zero cloud egress. Zero rate limits.
Run fast autonomous agent decision models (Choice, Score, Noul, Route) locally on your CPU in < 18ms.

🌟 Flagship Architecture

  • Pure-Rust SIMD Quantized Tensor Layer (src/tensor.rs): 4-way unrolled 8-bit quantized linear projection forward pass without C++ runtime dependencies.
  • Speculative Cascade Router (src/cascade.rs): Short-circuits high-confidence queries in <1ms and flags ambiguous prompts for deep fallback.
  • Online Preference Learning & Bandit Feedback (src/bandit.rs): Contextual UCB prior updates learning from real agent execution outcomes.
  • Fast Cosine Vector Context Index (src/vector.rs): Sub-millisecond embedding search for prompt states.
  • Embedded Native MCP stdio Server (local-jev mcp): Drop-in Model Context Protocol server for Claude Desktop, Cursor, and Hermes.
  • Micro-Batch Scheduler & Telemetry (/metrics): Prometheus metrics for real-time observability.
  • Official Client SDKs: Turnkey Python (local-jev) and TypeScript SDKs.

📦 Installation

cargo install local-jev

🧪 Verification & Benchmarks

local-jev v0.5.0: Flagship Multi-Language SDKs & MCP Server

Choose a tag to compare

@ronitgupta138 ronitgupta138 released this 10 Oct 17:20
4397214

⚡ local-jev v0.5.0 Production Flagship Release

Zero API keys. Zero cloud egress. Zero rate limits.
Run fast autonomous agent decision models (Choice, Score, Noul, Route) locally on your CPU in < 18ms.

🚀 New in v0.5.0

  • Native Stdio MCP Server: Drop-in Model Context Protocol server for Claude Desktop, Cursor, and Hermes (local-jev mcp).
  • Micro-Batch Scheduler & Telemetry: Coalesces concurrent agent queries and exports standard Prometheus metrics (/metrics).
  • Adaptive Model Backbone: Supports local neural weights files (ONNX/GGUF) alongside instant zero-weight heuristic inference.
  • Python Client SDK (local-jev): Clean synchronous client with single-pass choice and score methods.
  • TypeScript Client SDK (local-jev): Async typed fetch-based client for Node.js, Bun, and Deno.
  • Terminal Monitor & HUD (local-jev top): Live operational health dashboard.

📦 Installation

cargo install local-jev

🧪 Verification & Benchmarks

local-jev v0.1.0: 100% Offline Drop-In Alternative to TypeSafe Jev

Choose a tag to compare

@ronitgupta138 ronitgupta138 released this 10 Oct 17:07
a7ee36f

⚡ local-jev v0.1.0 Production Release

Zero API keys. Zero cloud egress. Zero rate limits.
Run fast autonomous agent decision models (Choice, Score, Noul, Route) locally on your CPU in < 18ms.

🚀 Features

  • Drop-in REST Compatibility: Seamless drop-in replacement for TypeSafe Jev API via export TYPESAFE_BASE_URL=http://localhost:8080/v1.
  • Single-Forward-Pass Logit Extraction: Zero autoregressive token decoding loops; evaluates decision probabilities in a single forward pass.
  • Constrained Softmax: Masks vocabulary exclusively to candidate option token IDs with temperature scaling.
  • Sub-Millisecond CPU Latency: p50 = 1.0µs benchmarked inference on AMD Ryzen CPU.
  • Self-Contained CLI: Includes serve, choice, score, route, and benchmark subcommands.

📦 Installation

cargo install local-jev

🧪 Verification & Health