Recent Posts
Zg Search AI Tool Review: How Alibaba’s ripgrep, BM25 & Vector Hybrid Beats Traditional Grep
5 hours ago 高效码农zg: Alibaba’s New Local Search Tool That Combines ripgrep, BM25, and Vector Search for Humans and Agents The first thing I did after installing zg (zvec‑grep), the newly open‑sourced tool from Alibaba’s Zvec team, was run this command in my terminal: zg query –human “An unseen creature left a few marks. What did the detective infer?” –limit 3 It returned the relevant passage from The Memoirs of Sherlock Holmes, complete with file name, line numbers, and offset. The same corpus also contained Alice’s Adventures in Wonderland, which is semantically related to “animal‑left marks” but didn’t make it into the top …
FireRedAudio: The All-in-One 9B Audio Model for ASR, TTS, and Understanding
1 days ago 高效码农FireRedAudio Hands-On: A Unified Audio Language Model for ASR, TTS, Editing, and Understanding Last month, I spent over three hours wrestling with a 40-minute interview recording. The workflow was standard but painful: run ASR to get a rough transcript, manually stitch timestamps to the content, switch to a standalone TTS model for voice cloning to re-record sections ruined by background noise, and then piece everything back together. FireRedAudio, recently open-sourced by the FireRed Team (affiliated with Xiaohongshu), collapses that entire pipeline into a single model. It shares a 9B-parameter LLM backbone across all tasks, using an Audio Encoder for understanding …
Why Software Engineering Fundamentals are Crucial in the AI Era
2 days ago 高效码农Why Software Engineering Fundamentals Matter More in the AI Era If you’ve used Cursor or similar tools lately, you know the drill: describe what you want, hit Tab a few times, and a feature works. You didn’t type every line. It’s faster than before. But here’s the question that creeps up later: when you no longer produce code line by line, how do you know your system won’t break in production? Andrew Ng recently laid out a perspective in his AI Engineering Skills Map: understanding how software works is becoming the dividing line between those who can effectively steer coding …
How We Cut AI Coding Costs 52% & Scaled 7x at Uber [Optimization Blueprint]
3 days ago 高效码农How We Slashed AI Coding Costs by 52% Per Session While Scaling Usage 7x at Uber By August 2026, our weekly active users for AI coding tools grew 7x, and agentic requests grew 9.4x. Yet, our total AI spend has been flat since April. The core strategy wasn’t negotiating lower API prices. It was eliminating zero-value token consumption while maintaining output quality. Here is exactly how we measured and optimized every layer of our AI Software Factory. The Four Layers of Our AI Software Factory We organize AI usage into four distinct layers. The higher the layer, the more control …
How to Use Ollama to Run Uncensored AI Models Locally with Cherry Studio
4 days ago 高效码农The Ultimate Guide to Cherry Studio: AI Workbench Mastery (Including Local Model Deployment) You’ve got your favorite AI models—GPT for drafting, Claude for polishing, DeepSeek for research. The hassle isn’t the models themselves; it’s the context switching. Copying prompts, re-uploading files, and jumping between browser tabs wastes more time than the AI actually saves. Cherry Studio fixes this. It’s an open-source desktop application that integrates APIs from GPT, Claude, DeepSeek, Kimi, and even locally-running models into a single, unified interface. In this guide, I’ll walk you through the exact setup process—from installation to local deployment—so you can stop wrestling with …
ContextPilot Review: How Tencent’s Toolkit Solves Long-Context AI Agent Chaos
5 days ago 高效码农ContextPilot Hands-on: Tencent’s Open-Source Solution for Long-Context AI Agents Long-context agents have a dirty secret: bigger windows don’t always mean better results. Run a multi-hop retrieval or cross-document QA for dozens of turns, and you’ll watch the working context bloat into a chaotic mess. The model starts forgetting what matters while holding onto irrelevant noise. Tencent just open-sourced ContextPilot to tackle this head-on. Instead of passively cramming more tokens into a fixed window, this framework teaches agents to actively manage their own context—deciding what to remember, what to forget, and when to reorganize. The code is available on GitHub, and …
Codex vs DeepSeek Harness: The 2027 Developer’s Choice (150 Benchmark Results)
6 days ago 高效码农Codex vs DeepSeek Harness: I Read Both Source Codes and Ran 150 Benchmarks So You Don’t Have To Two things happened in August 2026. OpenAI open-sourced the harness that powers Codex under Apache-2.0. DeepSeek released dsh v0.1 under MIT. Then the comparison articles started rolling in. The consensus was nearly identical across the board: Codex has a mature ecosystem, DeepSeek Harness is more flexible with its plugin architecture. That statement isn’t wrong. But it doesn’t help you make a decision. So I did the tedious thing. I pulled both codebases locally and read them line by line. Codex locked …
Building a Rendering Engine from Scratch: How Turtle’s History Cache Redefines Web Navigation
7 days ago 高效码农Building a Browser Rendering Engine from Scratch in Zig: Turtle’s Design and Trade‑offs Instead of caching the document, Turtle caches the already‑laid‑out fragment tree – so back navigation skips layout entirely. But it stops paying off on very large documents, and we measured exactly where that happens. Three rendering engines power the web: Blink, Gecko, and WebKit. Each is the work of large teams over many years. Can a single developer build a usable rendering engine from the ground up? And if so, which design choices make it feasible, and where are the unavoidable pitfalls? Turtle is a proof that …
TLive-Omni: How a 4B Open-Source AI Beats 30B Models for E-Commerce Live Streams
8 days ago 高效码农TLive-Omni: An Open-Source Omni-Modal Model for E-Commerce Live Streaming – 4B Outperforms 30B Live streaming presents a unique challenge for AI understanding. Information is scattered across multiple channels and cross-references itself constantly. The host says “this shoe has a hidden 4cm height boost” while flipping the sole toward the camera, and chat messages flash prices simultaneously. Single-modality models inevitably miss connections. Taobao & Tmall Group recently open-sourced TLive-Omni to tackle exactly this scenario. Two sizes are available – 4B and 9B – with model weights and a technical report now public. The benchmark results tell an interesting story: on core …
Inside DSH iOS Plugin: Run iOS Simulators & Real iPhones Directly in Your Chat
10 days ago 高效码农Deploying Tencent Hy4 Preview: vLLM vs SGLang Setup Guide + 1M Context Window API Tips
11 days ago 高效码农Deploying Tencent Hy4 Preview: vLLM and SGLang Setup Notes Tencent’s Hy4 preview is a 770B-parameter MoE model with 49B active parameters and a 1M context window. The weights are available on Hugging Face, ModelScope, GitCode, and CNB, including an FP8 quantized version. This guide walks through production deployment using vLLM or SGLang, plus the API calls you’ll need to actually use the model. Where to Find the Real Value in the README The official README is straightforward—architecture specs, benchmark numbers, license—but the most actionable content lives in the Inference and Deployment section. Both vLLM and SGLang are officially supported, and …
Fix Codex CLI unsupported_country_region_territory Error in Pi Agent: OAuth Proxy Guide
11 days ago 高效码农How to Configure Codex CLI in Pi Agent and Fix the unsupported_country_region_territory Error One of the more confusing issues when connecting Codex to Pi Agent is that the error message often points developers in the wrong direction. Many users immediately start modifying provider settings, model parameters, or configuration files. In practice, the problem usually appears much earlier in the authentication flow. A common example looks like this: Error: OAuth refresh failed for openai-codex OpenAI Codex token refresh failed (403): { “error”:{ “code”:”unsupported_country_region_territory”, “message”:”Country, region, or territory not supported”, “param”:null, “type”:”request_forbidden” } } At first glance, this appears to be a …
DeepSeek Harness Troubleshooting: How I Fixed the ‘deployment:persona already registered’ Error
12 days ago 高效码农DeepSeek Harness Error deployment:persona already registered: A Real-World Troubleshooting Journey from Reinstallation to .dsh Profile Conflicts Most DeepSeek Harness installation issues tend to revolve around model providers, API authentication, MCP connections, or tool integrations. This one was different. The error looked like a simple configuration problem at first glance: internal: resume failed for session “session-3836a0e6-9fe9-40d7-884e-876b3ebba209” Error: agent-presets: preset “standard” failed to mount failed to apply loader entry persona (@deepseek-ai/dsh-persona): prompt section “deployment:persona” is already registered The natural assumption is that a Persona configuration was duplicated somewhere inside the preset configuration. After several rounds of investigation, the root cause turned out …
How to Fix DeepSeek Harness Plugin Fails to Load: Missing dsh-sdk-protocol Error
13 days ago 高效码农DeepSeek Harness Plugin Fails to Load: Missing dsh-sdk-protocol and Platform Binary Warnings After installing @deepseek-ai/dsh-subagent-codex, running dsh web exits immediately with an error: Cannot find package ‘@deepseek-ai/dsh-sdk-protocol’. The plugin’s own node_modules doesn’t include it, and DSH won’t pull it in automatically. The Error: Plugin Loader Crashes During cordis:include The stack trace points to dsh-app-boot while building the plugin tree: Error: dsh: plugin tree failed to load: failed to apply loader entry include (cordis:include): failed to import loader entry subagent-codex (@deepseek-ai/dsh-subagent-codex): Cannot find package ‘@deepseek-ai/dsh-sdk-protocol’ imported from C:\Users\reanod\.dsh\profiles\web\node_modules\@deepseek-ai\dsh-subagent-codex\lib\index.js The critical path is …/dsh-subagent-codex/lib/index.js. That file contains an import statement referencing @deepseek-ai/dsh-sdk-protocol, …
How to Add Codex and Claude Code to DeepSeek Harness on Windows
14 days ago 高效码农How to Integrate Codex and Claude Code into DeepSeek Harness on Windows If you are trying to add Codex and Claude Code as subagents in DeepSeek Harness, the package names are now available from npm, but installing the packages alone is not enough. The main difficulty is keeping the DeepSeek Harness core packages, subagent packages, and session packages on compatible versions. This article documents the actual installation path and the errors encountered while integrating @deepseek-ai/dsh-subagent-codex and @deepseek-ai/dsh-subagent-claude-code into a DeepSeek Harness web profile on Windows. The key problem is version compatibility. In the example below, DeepSeek Harness itself is installed …
DeepSeek Harness Version Update: Why npm Version Stays the Same and How to Fix It
15 days ago 高效码农How to Update DeepSeek Harness When npm Version Doesn’t Change You ran npm update -g @deepseek-ai/dsh, saw a bunch of packages added and removed, then checked dsh -V and got the same version number back. This isn’t a user error — it’s how the @deepseek-ai/dsh package is published on npm. Why npm update Doesn’t Give You the Latest GitHub Release The npm next channel and GitHub Releases are published 「independently」. The GitHub Latest tag currently points to dsh-v0.1.0-rc.12, but the highest version of @deepseek-ai/dsh available on npm is 0.1.0-rc.7. That’s the root cause. The version you see on the GitHub …
How TeamAI Syncs AI Skills and Hooks Across Claude, Cursor, and CodeBuddy
16 days ago 高效码农How TeamAI syncs skills, hooks, and MCP servers across Claude Code, CodeBuddy, and Cursor Last month we plugged TeamAI into our team’s AI workflow. It fixed a few nagging problems we hadn’t found a good solution for: Each member had different versions of local skills and rules. The same request produced different results on different machines. Hook configurations relied on verbal sync. Someone forgot to install a security scanning script, and we almost committed a test‑environment secret. Onboarding new members meant manually copying configs to each local tool. Missing one step cost hours of debugging. TeamAI is not another AI …
Stop Just Reviewing Diffs: An Evidence-Based Approach to Evaluating AI Coding Agent Workflows
17 days ago 高效码农Better Hands-On Guide: How to Review AI Coding Agent Workflows with Evidence Reviewing Diffs is Not Enough: How to Evaluate AI Coding Agent Workflows AI coding agents write code quickly, but the workflows surrounding them are often fragile. If you only review the final code diff, you miss systemic issues. Better Harness is an open-source tool that analyzes the workflow behind the changes, gathers project and session evidence, and translates specific gaps into prioritized improvements. It focuses the analysis on a closed loop of feedforward guidance and feedback sensors. Feedforward guidance includes AGENTS.md, specs, and acceptance criteria. Feedback sensors include …
Professional Chinese Document Creation: How to Generate Properly Formatted Word Files from Markdown
18 days ago 高效码农Generate Properly Formatted Chinese Word Documents from Markdown: A Guide to typeset If you’ve ever tried pandoc x.md -o x.docx to create a Word document from Markdown, you know it works fine—for English blog posts. For Chinese contracts, agreements, service confirmations, proposals, or formal correspondence, the default output falls short in almost every way: paper size, fonts, line spacing, headers and footers, table breaks, heading widows, and signature blocks that drift across pages. Some of these issues only show up when you open the file in Word. typeset solves this. It turns Markdown into properly formatted Chinese .docx files, following …
How to Fix a Full 20GB Disk and Hermes Authentication Errors on Tencent Cloud
19 days ago 高效码农How to Troubleshoot a Full 20GB Linux Disk and Hermes Authentication Errors on Tencent Cloud Running Hermes on a Tencent Cloud server with a 20GB system disk can lead to several problems appearing one after another. In this case, the root partition first reached 100% usage, Hermes later stalled while checking a TTS dependency, the Gateway restart command reported a systemd user-session problem, and the mobile client eventually showed a Provider authentication failure. The useful part of this incident is that each stage produced concrete command output. That makes it possible to follow the troubleshooting process from disk usage analysis …