Top Picks

Self-hosted AI observability for tracking LLM requests, costs, latency, prompt traces, reasoning tokens, and PII masking across many model providers.

A cognitive memory engine for AI agents with temporal decay, contradiction detection, autonomous consolidation, knowledge graph support, and ANN recall via HNSW. The repo shows Rust and Python usage, MCP client setup, and a quick-start path for agents and developers.

A session manager for AI coding agents that lets developers run and monitor multiple agents in parallel from a TUI or web dashboard. It supports several agent frameworks, uses git worktrees and optional Docker sandboxes, and includes remote phone access plus PWA-style web use.

TypeScript SDK for building agent skills as typed state machines, with workflow steps, schema validation, branching, testing helpers, and packaging into distributable executables.

Kit is a developer workspace that combines an editor, Chromium browser, terminal, email, calendar, whiteboard, and autonomous agent with shared AI context across the whole environment.

A research-style demo showing LLM agents evolving through a genetic algorithm to play Pokémon in parallel sandboxed VMs. The page explains the setup, scoring loop, generations, results, and links to the repo and run artifacts.

WUPHF is a local-first AI agent workspace that runs a team of role-based agents with a shared wiki and persistent project memory. It supports Claude Code, Codex, OpenClaw, and local models via OpenCode, with install and build options shown on the page.

MCP server that lets AI coding agents work inside isolated dev containers using local Docker, DevPod, or GitHub Codespaces. It exposes tools for auth, workspace management, command execution, and container lifecycle handling.

A minimal Rust crate for building agentic workflows in applications, with ticket-based orchestration, tools, knowledge storage, schema validation, retries, budget policies, and multi-provider support.

A Codex browser plugin and MCP tool that captures live UI edits as structured implementation intent, with a local collector, installer scripts, and move-mode snapping for precise visual changes.

A CLI for structural code search and refactoring using AST patterns instead of regex. It supports find, replace, delete, and insert workflows, with examples for method renames, logging removal, and code inserts across files.

A pure-Java UI toolkit for building admin panels and internal tools, with AI-agent-friendly structures and examples for human-in-the-loop workflows.

A local control plane for coding agents that gates risky actions, records structured traces, and can replay sessions under different policies.

A self-hosted Git LFS server for BYOF setups, with support for GitHub, GitHub Enterprise, GitLab, and S3-compatible storage backends like Cloudflare R2, DigitalOcean Spaces, and Backblaze B2.

Skill pack for Claude Code that helps monitor, troubleshoot, and audit Kubernetes clusters using agent-friendly skills and cluster tooling.

Stage CLI is an AI-powered code review tool that groups local changes into logical chapters and helps users review what matters before diving into the code. It works with any AI agent, installs via npm, and can be invoked as a local command to open a browser-based review UI.

Local-first Mac vault for Markdown, JSON, and agent-ready context, with optional self-hosted sync, share links, and an HTTP API for agent workflows.

Flue is an open-source Python tool that lets coding agents control desktop software from the command line without MCP servers. The README says it works with apps like Photoshop, Blender, Unity, Houdini, and Microsoft Office, and supports harnesses such as Claude Code, Codex, Gemini CLI, Cursor, and OpenCode.

Sandboxed bash for agents with structured command output and full filesystem state capture after each run.

Runtime governance and containment for AI agents. Faramesh sits between an agent and its tools to enforce policy checks, approval steps, credential isolation, and tamper-evident audit logs before risky actions execute.

A Claude Code plugin that learns from user corrections and successful runs, then turns them into reusable rules and playbooks stored locally.

A ~1000-line C workflow runtime where an LLM can modify a DAG during execution using a restricted four-verb grammar, with every change written to an append-only audit log for replay and inspection.

A self-hosted TypeScript harness for building provider-neutral LLM agent systems inside an application or platform, with typed agent and workflow runtime, tool support, state, tracing, and example workspaces.

Herd is a lightweight desktop IDE for coordinating multiple AI coding agents at once. It supports parallel tasks, real-time status monitoring, and works with tools like Claude Code and Codex on Windows, macOS, and Linux.