ZCode: Z.ai's Open-Source Coding Agent Harness, From Turn Loop to Tool Dispat...
Most coding agents you can open up are a chat loop with a handful of tools bolted to its side. The interesting ones are the projects that treat the harness — the machinery that turns a model stream into safe,...
session-knowledge: Turning Claude Code History Into a Searchable Knowledge Ba...
If you have used Claude Code for more than a few weeks, you have felt the gap. Every session you have ever run is already on disk, under ~/.claude/projects/**/*.jsonl — every decision, every command that worked, every connection string you...
Reverify: Deterministic Grounding for AI Claims - Inside 2akouwu/reverify
Language models are fluent explainers and unreliable witnesses. Ask one to read a binary, audit a refactor, or describe a function it cannot see, and it will happily produce offsets, struct fields, and API names that simply do not exist...
OrcaReplay: Time Travel for AI Agents - Inside Continuum-AI-Corp/OrcaReplay
Your coding agent just deleted a file, “fixed” a test that is still failing, and exited 0. What do you do at that point? Today the honest answer is archaeology: scroll a terminal transcript, re-run and get a different failure,...
OpenChatX MCP: Turn ChatGPT into a Local Agent Runtime - Inside XiaoPuOuO/ope...
ChatGPT has become a remarkably capable planner, but for most of its life it has been a planner locked inside a browser tab. It can describe how to refactor your repository or run your test suite; it just cannot touch...
OpenBot: AI Coworkers With Their Own Computer - Inside CopilotKit/OpenBot
Most agent demos stop at the moment that matters: the model calls a tool, and the demo asks you to trust it. OpenBot, built by the CopilotKit team, is built around the opposite posture. Each AI coworker it hosts gets...
Open SEO MCP Skills: Free SEO Server + GEO Skills for Claude - Inside Ryze-AI...
Ask an “open-source SEO tool” where its numbers come from and the answer is usually the same: a paid SERP API behind the scenes, with the free part being little more than a dashboard on top. Ryze-AI-Adgent/open-seo-mcp-skills takes a different...
okf-agent-memory: Git-Native Persistent Memory for AI Agents - Inside okf-mem...
Every developer who has worked with an AI coding agent has felt the same frustration: the session ends, and the architectural decisions, domain discoveries, and hard-won operational facts evaporate with it. The common workarounds both fail in interesting ways. Stuffing...
NeoHorse: Recursive Self-Improvement via Agentic Post-Training - Inside Token...
Most agent models are trained once and then frozen, left to drift through whatever tool loops developers wrap around them. NeoHorse takes the opposite bet: the model itself sits inside a measurement loop, where a routing harness assigns tasks across...
NanoJev: A Nano Replica of Jev With Parallel Decisions and a Full Training Pi...
Most language-model agents decide by talking: they generate reasoning, then an answer token by token, then parse the text back into an action. NanoJev, a nano replica of the Jev decision model from TypeSafe, takes the opposite road. A state...
Magpie: Every Agent's Model, One Menu Bar - Inside yetone/magpie
Every coding agent ships with an opinion about who should serve its intelligence. Claude Code speaks the Anthropic Messages API and wants Anthropic keys, Codex wants OpenAI, Gemini CLI wants Google, and each one keeps its preference in a different...
ksor: One Governed Knowledge Record for Humans and AI Agents - Inside panaver...
Ask two AI agents in the same company a policy question and you can get two different answers — not because the models are weak, but because nobody ever defined which knowledge wins. The answers get assembled from an old...
Kev: Trainable Decision Models on Qwen - Inside jaredpalmer/kev
Ask a large language model a yes/no question and you get prose you have to parse. Ask it for a probability and you get a number that may not mean anything. Kev, by Jared Palmer, takes a different route: it...
Jev Review: Continuous Quality Review MCP Plugin - Inside NiazMorshed2007/jev...
AI coding agents are now good at producing code quickly, but the feedback they get back is usually binary: tests pass or they do not, the build compiles or it does not. What is missing is a cheap, structured signal...
Intent-Router: Typed IntentSpec Contracts for AI Agents - Inside angel291592/...
Hand an AI coding agent a request like “add caching to the user API” and one of two things happens. Either the agent guesses — it picks a cache backend, a TTL policy, and a failure behavior you never asked...
Human Atlas: An Interactive 3D Anatomy Explorer in the Browser - Inside ashem...
Interactive anatomy on the web tends to live at two extremes: heavyweight clinical platforms behind subscriptions, or static plates you can look at but never take apart. Somewhere in between sits a genuinely hard engineering problem — how do you...
Firecrawl: The Web Data API for AI Agents - Inside firecrawl/firecrawl
Every AI agent eventually hits the same wall: the web was built for browsers, not for language models. Give an agent a URL and it gets JavaScript boilerplate, cookie banners, nav menus, and anti-bot walls — none of which fit...
fast-jev-compaction: Jev-Scored Verbatim Compaction for Claude Code - Inside ...
Every long agent session eventually hits the same wall: the context window fills up and something must be forgotten. The conventional answer — asking a model to summarize the older turns into prose — has a structural flaw. A summary...
Easel: An Open-Source AI Agent for Social Media Creation - Inside ZJU-REAL/Easel
Most AI writing tools stop at the draft. They hand you text, maybe an image prompt, and then leave the hard parts - formatting for each platform, uploading, checking whether anything actually worked - to you. Easel, from the ZJU-REAL...
Concat: The Truly Free, Open-Source CapCut Replacement - Inside jub0t/Concat
Everyone who cuts short-form video knows the deal: the editor is free until you export, and then the watermark appears, or the 4K button asks for a subscription, or your footage quietly travels to somebody else’s cloud. Concat, at roughly...
Company Brain: A Teammate in Slack That Remembers Everything - Inside superme...
Most team knowledge never makes it into a document. It lives in Slack threads, gets decided in a hallway conversation that someone summarizes in a channel, and then evaporates the moment everyone scrolls past it. Search cannot save you, because...
Codenotch: macOS usage limits at the screen edge - Inside vinzdg/codenotch
Every AI coding assistant meters you differently. Claude Code counts a rolling five-hour session window plus a weekly cap, Cursor keeps its plan state in a local SQLite database, Codex reads its own auth.json, and none of them publishes a...
cf: The Agentic CLI for the Entire Cloudflare API - Inside cloudflare/cf
Cloudflare’s API surface is famously enormous — DNS, Workers, R2, KV, D1, Queues, Zero Trust, WAF, AI, and dozens of other products, each with its own endpoints and schema. Agents that try to drive that surface today have to guess...
Anthropic Skills: How SKILL.md Folders Teach Claude Specialized Work - Inside...
Every conversation with an AI assistant starts from zero. The model may know what a Word document is, but it does not know your company’s brand guidelines, your preferred redlining workflow, or the dozen gotchas that make Office XML corrupt...