Tag
Microsoft releases FrogNano-4B-2609, a compact repository-level coding agent derived from Qwen3.5-4B that is post-trained exclusively with reinforcement learning on ~1,500 synthetic SWE tasks via the TaskPilot pipeline and the lightweight five-tool Leaf harness, with no distillation from stronger models.
Redis creator antirez released ds4 (DwarfStar 4), a narrow C-based local inference engine for high-memory Mac, CUDA and ROCm machines, supporting DeepSeek V4/V4.1, GLM 5.x and Qwen3.8 Flash Next with asymmetric 2-bit quantization, SSD-backed KV cache, and CLI/server/agent interfaces under an MIT license.
The author benchmarks a local coding agent setup—Qwen3.8-27B Q4_K_M on a 24GB RTX 3090 via llama.cpp with OpenCode—reporting throughput figures, four bounded Python coding tasks (three passing checks), and a reasoning-budget failure where the model exhausted its 8K output allowance without producing a patch.
Earendil ships Pi 1.0, a hardened, minimal, extensible AI agent harness with features like Codemode (native MCP support), extension support for virtual models, and deferred tool loading, alongside the experimental Pi Durable package for long-running agentic applications. Both are MIT licensed and available today.
JenovaAI announced Ruby Coding Assistant, an AI agent that writes idiomatic, production-grade Ruby for scripts, gems, and Rails apps, with targeted fixes, backtrace diagnosis, and RSpec test generation.
作者从零实现了一个极简编码智能体，详细讲解了其架构：包含异步事件驱动的 ReAct 代理循环、带沙箱的 read/write/edit/bash 工具、支持分支与回溯的 JSONL 会话存储，以及用 TypeScript 编写并通过 stdin/stdout JSON 通信的 TUI。
Qwen3.8-27B-pi is a Qwen3.8-27B fine-tune built for the Pi coding agent harness, trained with SFT on successful agent sessions followed by GRPO with a success-conditioned reward that enforces effort-ordered reasoning. It matches base xhigh performance on Terminal-Bench 2.1 at medium effort while using ~41% fewer output tokens, and is released in BF16, FP8 and 17 GGUF formats.
ANUS 是一款开源编码代理，仅使用免费 LLM 运行：它维护一个免费模型排名列表，自动将请求路由到当前可用的最佳模型，当某个模型达到速率限制时自动切换到下一个备选模型。项目采用 MIT 协议，在 GitHub 上获得超过 6000 星。
Devin's latest update makes the AI software engineering agent 30-70% cheaper depending on mode, achieved by integrating the latest models like SWE-2 and optimizing its harness through batched tool calls and improved prompt caching, while Devin Fusion now leads FrontierCode 1.1 Extended at $0.60 per task.
The AI SDK has reached 30M weekly downloads, showcasing fx, a minimal and performance-focused coding agent written in Zig with WebAssembly support.
A researcher demonstrated successful memory poisoning attacks on an AI agent's memory server, achieving a 100% success rate, leading to fixes in Knowl 5.24.0, though some vulnerabilities persist.
CoreWeave ARIA is an AI-powered coding agent integrated into Weights & Biases that automates the entire research loop, from hypothesis formation to experiment execution and analysis.
Cline's desktop app has become the #2 top coding agent on OpenRouter after its release, trending on the platform.
A developer ports a mini web browser to Rust and Swift for improved speed and security, and introduces 'fx' as a tiny, native coding agent tool.
The paper proposes AutoRef, a method that automatically optimizes the executable harness of an image-generation agent using a coding agent that rewrites harness code while keeping models frozen. The resulting AutoRef-Harness improves open-weight FLUX.2[klein] 4B from 5.72 to 7.37 on held-out four-reference MultiBanana tasks, matching or exceeding proprietary models like NanoBanana Pro and GPT-Image-1.5, and transfers across generators, benchmarks, and reasoning models without re-optimization.
jevgrep is a research agent CLI powered by jev from typesafeai that reduces coding agent costs by 40% by providing code context through natural language questions.
Drawgent is a Rust-based tool that connects coding agents like Claude Code, Codex, or opencode to an Excalidraw canvas for live diagram editing and collaboration.
PiG is a Go port of the Pi coding agent harness, providing faster startup and lower memory usage as a single static binary with support for live-reloaded extensions.
Microsoft introduces Coding-Agent Skill Distillation (CASD), a prompt optimization method where an off-the-shelf coding agent analyzes agent logs to write optimized prompts in one pass, outperforming previous techniques like GEPA and SkillOpt at a lower cost.
An open-source coding agent harness was stress-tested by using at least 100 AI agents to update its own documentation, demonstrating efficient parallel task handling and identifying real issues in the docs.