These sources are tools and setups built around coding agents. Most are project pages and launch threads. This page lists 24 sources, oldest first where dated.
Windsurf vs Cursor: which is the better AI code editor?
Sewell compares two editors feature by feature from personal use. The post is about 9 minutes long. It also promotes the publisher's own visual editing extension.
Show HN: I built an MCP to help build Cursor for Video Editing
The poster lists the tools of a Python MCP server for editing video and audio. The post is short and the thread has no comments. Its only evidence is the tool list and a claim of automated tests.
Why Go is a good fit for agents
The post comes from the CEO of Hatchet, which sells an orchestration platform for agents. It argues from Go's concurrency features and the author's experience. The evidence is a toy code example and observations of customers. The post also lists reasons not to pick Go.
Rethinking CLI interfaces for AI
The author draws on an IDA Pro MCP server and Claude Code use. The evidence is anecdotes and a few code snippets. The proposals are sketches, with no systematic tests.
Porrima — agent system
The page lists features of a project by one named author built on Llama.cpp. It covers memory design, hardware needs and recommended models as of September 2026. It offers feature claims and no benchmarks.
omp — a coding agent with the IDE wired in
This is a landing page with a tagline, an install command and little else. It gives no benchmarks or documentation text. The page is very short, so it offers claims only.
Show HN: I built a tool to create AI agents that live in iMessage
The poster pitches Flux, which turns a described personality into an iMessage contact. Commenters report their own tests, ask how it works and one questions Apple's rules. The evidence is anecdote and one reading of the public repo.
CLI agents make self-hosting on a home server easier and fun
The thread discusses Jordan Fulghum's January 2026 post on a Claude Code home server. Debate centers on Tailscale, exposed ports and trust in a closed AI service. The evidence is personal experience from the author and many commenters.
My Homelab AI Dev Platform
The linked post is dated 14 June 2026. It describes a homelab where OpenCode makes changes and the author approves pull requests. The thread has 367 points. Commenters share their own agent setups and ask about models, GitOps and cost.
Show HN: Pulse – Dashboard for Claude Code, approve tool calls from your phone
A student presents Pulse, which shows a Claude Code session and relays approval prompts. Replies mostly point to the built-in remote control feature. The evidence is the poster's description and comments about the built-in feature's reliability.
Show HN: Mindwalk – Replay coding-agent sessions on a 3D map of your codebase
The creator presents a local Go tool that draws a repository in 3D. It replays agent sessions on that map. The thread has 162 points. Commenters praise the idea and press the creator on its use case.
How to stop Claude from saying load-bearing
The post treats Claude's repeated phrases as a nuisance. It offers a script that rewrites them on screen. It is tagged as parody. The only evidence is a script and a settings snippet.
doobidoo/mcp-memory-service: MOVED to github.com/doobidoo/mcp-memory-service - archive only, no further development
The page is the README of mcp-memory-service, a memory server for AI agents. A notice says the project moved to GitHub on 5 September 2026. The README offers benchmark tables, user quotes and a release history. The benchmarks are the project's own.
A2UI
The page introduces A2UI, a protocol for agents to describe interfaces in a declarative format. It lists spec versions, features, a quickstart and two demo videos. It is project documentation, so its claims are the project's own.
neckbeard
The README describes a plugin that splits infrastructure work between agent and Go tool. It is unusually frank about what is untested. It includes a cost table, a status table and a benchmark summary. The evidence is the author's own.
trollbridge — let your agents run amok
The page explains a policy, advisor and manual-approval flow for agent web traffic. It includes a quickstart, an FAQ and a comparison table. It offers design claims and one testimonial.
Graphicle
The README describes a Rust port of Graphify plus a memory layer. It serves agents over MCP. It is dense with benchmark figures against Graphify and memory systems. All were run by the company. The README says it ties grep on finding code and does not claim token savings.
palmier-io/palmier-pro: macOS video editor built for AI
The README is short. It describes a Swift video editor with built-in AI generation and a local MCP server. A notice says releases after v0.7.6 are proprietary. The repository keeps the older GPL source. The page gives no tests or demonstrations.
Changelog | Polytoken
The page is the release notes of Polytoken, a terminal-based coding agent. It covers versions 0.2.0 through 0.9.7 in reverse order. The notes are written in a jokey first-person voice. It records changes. It does not test how well the tool works.
Slopo
The README describes Slopo, which finds code doing the same job in different ways. The tool's own author wrote it. It gives a usage guide and design notes but no results of its own. The benchmark and example reports sit on linked pages.
Dirac, an Open-source AI coding agent for efficiently doing complex work
The README presents Dirac as a fork of Cline built for long tasks. It lists features and a benchmark of eight refactoring tasks against six other agents. The authors ran the benchmark themselves. They say Dirac finished all eight at the lowest average cost.
Vibium
The README presents Vibium as a verification layer for coding agents. It shows a command-line interface, an MCP server and client libraries for three languages. The page offers command examples and an architecture diagram. It gives no test results or comparisons.
Worktrunk — Git worktree management for parallel AI agent workflows
The page introduces the tool and compares its commands with plain git. It also shows install steps. It is product documentation with sample terminal output. It gives no independent tests.
How to set up your spare Mac for Claude Code to fully control - a step-by-step guide
The guide gives 16 numbered steps with shell commands for SSH, sleep and remote access. It rests on the author's own setup. It links helper scripts and a newsletter.
Glossary
- ACP
- Agent Client Protocol, a standard that lets editors host coding agents.
- Agent mode
- A chat mode where the AI picks files and runs commands itself.
- Allow list
- A list of destinations that are permitted.
- Apache 2.0
- A permissive open-source license that allows broad reuse.
- ARIA role
- A label in a web page that says what kind of control an element is.
- AST
- Abstract syntax tree, the structured form of code that tools use to understand it.
- Catalog
- The approved list of components an agent may ask the client to draw.
- CGNAT
- A setup where many homes share one public address, which blocks inbound connections.
- Changelog
- A list of the changes made in each release of software.
- Channel
- A Go pipe that lets concurrent tasks pass messages to each other.
- Checkov
- A scanner that checks infrastructure code against security policies.
- Claude Code
- Anthropic's command line coding agent.
- CLI
- A command-line interface, used by typing commands.
- Cline
- A VS Code extension that tries steps, checks results and repeats.
- Code graph
- A map of a program's files, functions and calls, and how they link.
- Compaction
- Shrinking a long chat history so it fits the model's memory.
- Context compaction
- Summarising a long conversation so it fits within the model's memory limit.
- Context window
- The amount of text a model can consider at one time.
- Context.Context
- A Go standard object used to signal that running work should stop.
- Cosine similarity
- A score from -1 to 1 for how close two embeddings are.
- Cursor
- An AI code editor, a fork of VS Code, with a Composer chat mode.
- Daemon
- A background program that keeps running and serves the interface.
- Declarative UI
- A description of what an interface contains, not code that builds it.
- Embedding
- A list of numbers that represents the meaning of a text, used for similarity search.
- Facet
- In Polytoken, a named mode such as plan or execute with its own instructions.
- Fargate
- An AWS service that runs containers without managing servers.
- Forgejo
- A self-hosted Git service.
- GGUF
- A file format for storing models that Llama.cpp can run.
- Git worktree
- A Git feature that checks out several branches into separate folders at once.
- GitOps
- Deploying changes automatically from a Git repository.
- Goroutine
- A very light unit of concurrent work in Go.
- Gource
- An older tool that animates a project's history as a growing tree.
- GPLv3
- A free-software licence that requires shared source for modified versions.
- Graphify
- An open-source code-graph tool that Graphicle's engine is ported from.
- Harness
- The software wrapper that lets a language model run tools and edit code.
- HCL
- The configuration language used by Terraform and OpenTofu.
- Homelab
- A set of servers kept at home for learning and personal services.
- Hook
- A script that runs automatically at a set point in a program's work.
- HTTP proxy
- A program that sits between a client and the web and forwards its requests.
- IMessage
- Apple's messaging service built into iPhones and Macs.
- Information architecture
- The practice of deciding how information is organized and shown.
- Infracost
- A tool that estimates the monthly cost of infrastructure code.
- JSONL
- A text format with one JSON record per line.
- Knowledge graph
- A map of facts stored as items joined by labelled links.
- LaunchAgent
- A macOS item that starts a program automatically in a user's session.
- LiteLLM
- A library that lets one program call many AI model providers.
- Llama.cpp
- An open program for running language models on local hardware.
- LoCoMo
- A benchmark that tests how well systems remember long conversations.
- LongMemEval
- A benchmark that tests whether a system recalls details from long chat histories.
- MCP
- Model Context Protocol, a standard way for AI tools to call outside services.
- MCP server
- A program that offers tools to AI clients through the Model Context Protocol.
- MessageDisplay hook
- A Claude Code hook that can change a reply's text before it is shown.
- OIDC federation
- A way for a pipeline to prove its identity without storing long-lived cloud keys.
- OpenCode
- An open source coding agent with a built-in web interface.
- OpenTofu
- An open-source tool for describing cloud infrastructure in code files.
- Orchestration
- Software that schedules, runs and tracks many jobs.
- Passwordless sudo
- A setting that lets a user run admin commands without typing a password.
- Pre-commit hook
- A script that Git runs before accepting a commit.
- R@5
- Recall at 5: how often the right answer appears in the top five results.
- Relay
- A server that passes messages between two systems.
- Remote control
- A Claude Code feature for answering prompts from the Claude phone app.
- Renderer
- Client code that turns the description into real on-screen widgets.
- Reranker
- A model that reorders search results by relevance.
- Rust
- A programming language often used for fast, low-level software.
- Sandbox
- An isolated environment that limits what a program can touch.
- Self-hosting
- Running services such as photos or passwords on a server you control.
- Session log
- A file that records what a coding agent read, searched and edited.
- Show HN
- A Hacker News post type where makers present their own work.
- Skill
- Saved instructions that a coding agent can run when asked.
- Smithery
- A site that lists MCP servers.
- SSH
- A secure way to log in to and run commands on another computer.
- Subagent
- A second agent that a main agent starts to do part of a task.
- SWE-bench Verified
- A set of real software issues used to test coding tools.
- Systemd unit
- A config that makes Linux start and manage a service.
- Tailscale
- A service that links your devices over a private network.
- Timeline
- The strip in a video editor where clips are arranged in time.
- TLS interception
- Decrypting secure traffic at the proxy so it can be inspected.
- Tmux
- A terminal program that keeps sessions running after you disconnect.
- Tool call
- A request from an AI agent to run an action, such as editing a file.
- ToS
- Terms of service, the rules a company sets for using its product.
- Trajectory
- The ordered path of steps an agent takes through a task.
- Tree-sitter
- A parser that turns source code into syntax trees without a model.
- Treemap
- A chart that shows a folder tree as nested rectangles.
- TUI
- A text-based user interface that runs inside a terminal.
- Vibe coded
- Written mostly by prompting an AI model.
- WebDriver BiDi
- A standard protocol that lets programs control a web browser in both directions.
- Windsurf
- An AI code editor, a fork of VS Code, with an agent called Cascade.
- WireGuard
- An open VPN protocol that Tailscale builds on.