AI agents

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 CLI tool that gives coding agents names from a roster so teams can track who is working on what across Claude Code, Codex, tmux, and cloud sessions.

A local LLM gateway that redacts secrets before requests reach model providers and restores them locally afterward. It is aimed at agent workflows where prompts and tool calls may contain real credentials.

A macOS Chrome relay that lets coding agents like Claude Code, Codex, Cursor, and Paseo use your signed-in Chrome in background tabs without taking focus.

Premortem is an AI red-teaming tool for startup ideas. Users paste an idea, six specialized agents critique it, and a seventh agent synthesizes the findings into a memo.

HEIDES is a deterministic local harness for AI coding agents that maps a codebase, grounds plans against the current graph, and checks proposed patches for conflicts, edge cases, and security issues before changes land.

Octomind Routines is a scheduled-agent product for running agents on persistent cloud machines with visible run history, state files, cost controls, and timezone-aware scheduling.

A lightweight shell-based guard that blocks destructive commands before an AI coding agent can run them. It targets risky actions like git reset --hard, rm -rf, force-pushes, and destructive cloud/database commands, with local regex rules and a test script.

A small macOS menu bar app that keeps your Mac awake so long-running AI agents can continue running with the lid closed.

A CLI for running untrusted MCP servers in sandboxed containers with deny-by-default egress, traffic watching, and host approval workflows. It can also compose servers into agents and expose them over OCI registries.

Authoryze is an MCP-native payment layer for AI agents that issues single-use virtual cards for approved purchases and adds spending limits, merchant rules, approvals, and audit trails.

TaskPeace is an MCP-native task manager for AI coding agents. It gives agents one ranked queue to pull from, with a live cockpit for humans to monitor work, and supports Claude Code, Cursor, ChatGPT, and other MCP or REST-based agents.

An open-source AI agent platform for production workflows, with isolated containers, credential proxying, and support for 100+ business and productivity integrations.

Google’s announcement for a built-in computer use tool in Gemini 3.5 Flash. The page explains that developers can use it to build agents that interact across browser, mobile, and desktop environments, with safety features and access through the Gemini API and Gemini Enterprise Agent Platform.

A black-box benchmark report on how AI-generated tests detect functional bugs in live APIs across 20 scenarios and 7 systems.

arXiv paper describing QUEST, an open family of deep research agents from 2B to 35B parameters, plus a synthetic-task training recipe and released models, data, and scripts.

HookGuard is a CLI security scanner for AI coding agent configuration files. It looks for malicious hooks, invisible Unicode, credential exfiltration patterns, and prompt-injection text in files like CLAUDE.md, AGENTS.md, Cursor rules, and GitHub Copilot instructions.

Aion is a collaborative coding game where AI agents propose code diffs to a live game, vote on changes, and ship the winning diff server-side. It includes an MCP endpoint so agents can connect, read the rules, clone the repo, and submit proposals.

A local CLI for Claude Code users that turns repeated corrections into evidence-backed CLAUDE.md rules, with a curated rule pack mined from Claude Code issue history.

A CLI code review tool that inspects an agent’s session, not just the diff, to check whether it did what you asked, what it decided on its own, and whether its claims match the code.

A design write-up and product page for AgentTerm, a terminal-based AI workspace focused on steering agents, reviewing code, opening documents in-window, and coordinating work across desktop and devices.

A local-first Rust runtime and IDE for running and governing AI agents with human approval, capability-based permissions, and audit logging.

HDTP is a protocol specification for agent-to-agent communication on behalf of people, centered on certificates, contact approval, per-caller permissions, and sealed end-to-end calls.

A long-form research article on why browser-based AI agents fail on real websites and how accessibility engineering and task-level tests can reveal the breakpoints.