Releases·Tool Release
I built a local AI agent with a headless browser and sandboxed shell — plug-and-play skills, no cloud required
A developer released Agent + Skill Store, a local AI agent installed via pip that drives headless Chromium, runs shell commands in a sandbox, and loads capabilities as folder-based 'skills'. It works with any OpenAI-compatible API and ships with a self-hostable skill registry. The post is the release announcement, so practitioner relevance is that it offers a transparent, inspectable alternative to cloud-hosted agent runtimes.
7 engagement1 sourceTue, Sep 29, 2026, 01:51 AM
AI Overview
- Reddit user releases Agent + Skill Store, an MIT-licensed local AI agent (pip install agent-code, Python 3.10+) with headless Chromium via DrissionPage and a sandboxed shell with allowlist and workspace jail
- Skills are drop-in folders with skill.py and skill.md exposing SKILL, TOOL_SCHEMAS, and TOOL_CALLABLES; bundled skills include filesystem, public_api (700+ free APIs), and temp_mail
- Works with any OpenAI-compatible API (OpenAI, Together, Groq, DeepSeek, local vLLM), with multi-key rotation, chat auto-save, and Stop/Continue buttons that resume from the last tool result
- A self-hosted GPL-3.0 skill registry server (skills-manager.freesrv.com) lets users publish and install skills via GUI; author warns skills are arbitrary Python and recommends Docker for untrusted use
Related
Releases·Tool Release
Developer launches OpenHub TUI for discovering and installing AI agent skills
A developer built OpenHub, a terminal-based tool that indexes and installs AI coding agent skills, prompts, and MCP servers from a central repository. It exports skills as SKILL.md files and handles binary/MCP server setup, aiming to eliminate manual configuration.
A developer built OpenHub, a terminal-based tool that indexes and installs AI coding agent skills, prompts, and MCP servers from a central repository. It exports skills as SKILL.md files and handles binary/MCP server setup, aiming to eliminate manual configuration.
0 engagement1 sourceredditFri, Jul 24, 2026, 06:03 AM
Products·launched
Local Python sandbox for LLM agents (E2B alternative)
Vivarium is a local, single-user Python sandbox for data analytics that serves as an alternative to E2B. It allows LLM agents to run Python code safely without granting full bash access, addressing security concerns. Users install it as a skill for their favorite agent via a SKILL.md file.
Vivarium is a local, single-user Python sandbox for data analytics that serves as an alternative to E2B. It allows LLM agents to run Python code safely without granting full bash access, addressing security concerns. Users install it as a skill for their favorite agent via a SKILL.md file.
0 engagement1 sourceredditTue, Jul 28, 2026, 10:05 AM
Products·launched
On-device AI agent launcher for Android phones
A personal AI agent that runs entirely on a phone, replacing the launcher interface. It uses LLMs to process commands and manage data locally, eliminating the need for external servers. Built for users who want privacy and offline capability.
A personal AI agent that runs entirely on a phone, replacing the launcher interface. It uses LLMs to process commands and manage data locally, eliminating the need for external servers. Built for users who want privacy and offline capability.
11 engagement1 sourceredditSun, Aug 2, 2026, 11:02 AM
Products·building
Memorall: open-source local-first browser agent workspace
Memorall is an open-source, local-first agent workspace that runs in the browser. Users describe an agent in plain English, and Memorall turns that into an editable agent with instructions, capabilities, tools, skills, memory, and execution flow. It solves the problem of browser AI tools being limited to chat boxes by providing a full agent harness.
Memorall is an open-source, local-first agent workspace that runs in the browser. Users describe an agent in plain English, and Memorall turns that into an editable agent with instructions, capabilities, tools, skills, memory, and execution flow. It solves the problem of browser AI tools being limited to chat boxes by providing a full agent harness.
1 engagement1 sourceredditFri, Aug 21, 2026, 08:20 AM
Products·launched
AgenticBrowser — a CLI-first agentic browser that works with your existing AI subscription
A CLI-first agentic browser that automates browsing tasks using LLM agents. It connects to AI subscriptions the user already pays for (OpenAI, Anthropic) or local models, avoiding extra search API costs. Built for developers frustrated with crashing alternatives like BrowserOS, with all data kept local.
A CLI-first agentic browser that automates browsing tasks using LLM agents. It connects to AI subscriptions the user already pays for (OpenAI, Anthropic) or local models, avoiding extra search API costs. Built for developers frustrated with crashing alternatives like BrowserOS, with all data kept local.
31 engagement1 sourceredditSat, Sep 5, 2026, 11:34 AM