I am building anyrouter.dev
I am building AnyRouter to solve my problem of collecting free LLM tokens from everywhere and unifying the MCP Gateway. Yes, it's another router, but this is a router of routers.
Tôi là Duyệt
Sr. Data Engineer. Rustacean at night
Latest posts
Open Managed Agents
I'm building Open Managed Agents (OMA) in the open — a self-hosted control plane on Cloudflare that drives agents on any sandbox backend from one UI and one API, and stays compatible with the Claude Managed Agents API. Here's where it stands.
Agent Sandbox on Kubernetes
I run Claude inside agent sandboxes on a home Kubernetes cluster. One consumes issue tickets into PRs, another triggers on Sentry webhooks to diagnose and patch runtime bugs. Here is the bigger loop, and where the new Kubernetes Agent Sandbox CRD fits.
Goal and Loop
You've probably heard of the agent loop by now. This is just a quick note on how I actually use it day to day — a `goal` and a `loop` — to get a lot more shipped.
Cowork as Planner
Using Claude Cowork as vision planner and market research
Autonomous Claude Code Migrate Nextjs to Tanstack Start
The second ultrawork run: I asked Claude Code to migrate chmonitor.dev from Next.js to TanStack Start and went to sleep. It planned the work, spawned parallel agents, and kept going for days.
Letting Claude Code work overnight, 2026
A quick snapshot of where coding agents are in 2026: you can hand one a real, multi-part task, walk away, and come back to merged PRs, a live deploy, and a few issues it worked around on its own.
Cloudflare is all you need
Cloudflare's Agents SDK runs stateful TypeScript agents on Durable Objects, with Workers AI, AI Gateway, AI Search, Vectorize, Browser Run, Queues, Workflows, Email, Dynamic Workers, and Sandboxes around it. Plus Flue, the open agent harness from the Astro team Cloudflare acquired.
Claws
OpenClaw, NanoBot, PicoClaw, ZeroClaw, NullClaw, NemoClaw, ...
Claude Code
Why Claude Code is still the king among coding agents I've tried
Claude Code Setup
How I configure Claude Code - auto-compact, permissions, MCPs, and skills
Claude Design
Claude Design reads your codebase and builds a design system from your colors and typography
CLAUDE.md, AGENTS.md
Keep your coding agent's project instructions consistent across sessions
Coding Agents
Reflect on what I'm thinking and doing in this LLM era
duyet/claude-plugins
A collection of plugins, skills, MCPs, commands, and hooks I use for Claude Code
Interview Mode
Ask clarifying questions before planning complex tasks
Long-running and self-improving coding agent
Build a long-running autonomous coding agent that runs 24/7
opencode
A coding agent with nice UI/UX that consolidates all your subscriptions
Parallel Sub-Agents / Teams
Lead a team of parallel agents with leader delegation to senior and junior roles
Plan Mode
Plan mode performs significantly better than just prompting directly
z_claude, mi_claude & or_claude
Wrapper scripts to use Claude Code with alternative providers
ClickHouse Rust UDFs
In Data Platform System with ClickHouse, rather than extracting data from ClickHouse for processing in external systems, we can perform transformations directly within ClickHouse itself. ClickHouse can call any external executable program or script to process data. My idea is using custom **User-Defined Functions (UDFs) written in Rust** to handle data transformations between tables.
ReplicatedReplacingMergeTree
Learn how to set up and manage ReplicatedReplacingMergeTree in ClickHouse on Kubernetes. This comprehensive guide covers cluster setup with ClickHouse Operator, data replication, performance tuning, and best practices for high availability deployments.
ReplacingMergeTree
My favorite ClickHouse table engine is `ReplacingMergeTree`. The main reason is that it is similar to `MergeTree` but can automatically deduplicate based on columns in the `ORDER BY` clause, which is very useful.
MergeTree
After starting this series ClickHouse on Kubernetes, you can now configure your first single-node ClickHouse server. Let's dive into creating your first table and understanding the basic concepts behind the ClickHouse engine, its data storage, and some cool features
Monitoring ClickHouse on Kubernetes
Complete guide to monitoring ClickHouse on Kubernetes. Learn about built-in dashboards, Prometheus + Grafana setup, powerful system tables for monitoring queries, and the ClickHouse Monitoring UI dashboard. Includes practical examples, essential monitoring queries, and best practices for production observability.
ClickHouse SELECT Advances
Dynamic column selection (also known as a `COLUMNS` expression) allows you to match some columns in a result with a re2 regular expression.
ClickHouse on Kubernetes
Complete guide to deploying ClickHouse on Kubernetes using the Altinity ClickHouse Operator. Learn how to set up your first single-node cluster, configure persistent storage, manage users, and customize ClickHouse versions. Includes practical examples and best practices from production experience managing clusters with trillions of rows.
2023 - A Year of Moving
So, I hadn't really planned on writing a summary post for this year, as lazy as I am, but somehow, here we are.
Rust: Processing JSON with serde
Serialize and deserialize JSON in Rust using serde and serde_json — derive macros, field attributes, enum representations, file I/O, dynamic values, and error handling.
Apache OpenDAL in Rust to Access Any Kind of Data Services
OpenDAL is a data access layer that allows users to easily and efficiently retrieve data from various storage services in a unified way such as S3, FTP, FS, Google Drive, HDFS, etc. They has been rewritten in Rust for the Core and have a binding from many various language like Python, Node.js, C, etc..
My Neovim Setup in 2023
It's been years since I first started using neovim and I've been updating it regularly ever since.
DuckDB
In this post, I want to explore the features and capabilities of DuckDB, an open-source, in-process SQL OLAP database management system written in C++11 that has been gaining popularity recently. According to what people have said, DuckDB is designed to be easy to use and flexible, allowing you to run complex queries on relational datasets using either local, file-based DuckDB instances or the…
Airflow control the parallelism and concurrency (draw)
How to control parallelism and concurrency
Fossil Data Platform Rewritten in Rust 🦀
My data engineering team at Fossil recently released some of Rust-based components of our Data Platform after faced performance and maintenance challenges of the old Python codebase. I would like to share the insights and lessons learned during the process of migrating Fossil's Data Platform from Python to Rust.
Running Spark in GitHub Actions
This post provides a quick and easy guide on how to run Apache Spark in GitHub Actions for testing purposes
Migrate (again) to Next.js 13 Pro Max
I've been using Gatsby since 2019 and while it's great that I can write my blogs in NeoVim, commit and push to Github, and have Cloudflare Pages build and publish to CDN, but I was becoming frustrated with the slow building times and the overall maintenance requirements.
Why does Helm Charts interpret 0777 as 511?
Why does Helm Charts interpret 0777 to 511? It took me quite some time to debug it.
GPT vs Traditional NLP Models
The field of Natural Language Processing (NLP) has seen remarkable advancements in recent years, and the emergence of the Generative Pre-trained Transformer (GPT) has revolutionized the way NLP models operate. GPT is a cutting-edge language model that employs deep learning to generate human-like text. Unlike conventional NLP models, which required extensive training on specific tasks, GPT is…
Ask ChatGPT about 20 important concepts of Apache Spark
I asked ChatGPT to explain 20 important concepts of Apache Spark. Let's see what it has to say!
Rust Data Engineering: Processing Dataframes with Polars
If you're interested in data engineering with Rust, you might want to check out Polars, a Rust DataFrame library with Pandas-like API.
Data Engineering Tools written in Rust
This blog post will provide an overview of the data engineering tools available in Rust, their advantages and benefits, as well as a discussion on why Rust is a great choice for data engineering.
Why ClickHouse Should Be the Go-To Choice for Your Next Data Platform?
Recently, I was working on building a new Logs dashboard at Fossil to serve our internal team for log retrieval, and I found ClickHouse to be a very interesting and fast engine for this purpose. In this post, I'll share my experience with using ClickHouse as the foundation of a light-weight data platform and how it compares to another popular choice, Athena. We'll also explore how ClickHouse can…
Airflow Dataset (Data-aware scheduling)
Airflow since 2.4, in addition to scheduling DAGs based upon time, they can also be scheduled based upon a task updating a dataset. This will change the way you schedule DAGs.
Cargo: Patch Dependencies
There are several scenarios when you will need to override or patch upstream dependencies. Like testing a bugfix of your crates before pushing to crates.io, a non-working upstream crate has a new feature or a bug fix on the master branch of its git repository that you'd want to try, etc. In these cases, the [patch] section of Cargo.toml might be useful.
Cargo: workspace inheritance
Since 1.64.0, Cargo now supports workspace inheritance, so you can avoid duplicating similar field values between crates while working within a workspace. Workspace inheritance can include things like shared version numbers, repository URLs, or rust-version.
Rust: Why ? is good
In Rust, the question mark (?) operator is used as an alternate error propagation method for functions that yield Result or Option types. The ? operator is a shortcut that minimizes the amount of code required in a function to quickly return Err or None from the types Result , or Option.
Rust: indoc
indoc là một crate nhỏ nhưng hữu ích giúp canh lề (indented documents). indoc!() macro nhận multiline string và un-indents lúc compile time, xoá tất cả khoảng trắng đầu tiên trên cách dòng dựa theo dòng đầu tiên.
Rust: Rayon - A data parallelism library for Rust
rayon là thư viện data-parallelism cho Rust, gọn nhẹ và dễ dàng convert từ code tính toán tuần tự sang song song mà vẫn đảm bảo không lỗi data-race.
Rust automatically upgrade to a new edition
Every two or three years, the Rust team produces a new Rust edition. Each edition contains a lot of changes. Each edition brings together the features that have landed into a clear package with fully updated documentation and tooling.