Developer tools and AI agents, reviewed from the source. Each review records what was built and run locally, separately from what the product claims.
Clean-room Photoshop rebuilt in pure Rust, scriptable over CLI and MCP.
Desktop studio for 100+ databases with RBAC and agent-safe MCP access.
Personal agent that works inside its own sandboxed Docker desktop.
Native macOS terminal for running coding agents side by side in separate worktrees.
Scientific paper search for agents that abstains instead of guessing.
Shared system of record where stateless agents propose changes instead of overwriting them.
Self-hostable multiplayer data workspace where agents' queries and charts stay inspectable.
Desktop workspace where several coding agents work as persistent teammates.
Native desktop app pairing per-project terminals with a kanban board that runs agents.
Local review deck for pruning Claude Code's persistent memories.
Control plane that runs coding agents on remote machines in isolated worktrees.
In-repo task board and MCP server that you and your coding agents share.
Local web hub that runs Claude Code, Codex and OpenCode as reusable bots.
Open-source CLI that audits AI agents with a judge model and grades them A to F.
Local proxy and hooks that police coding-agent tool calls and estimate their cost.
Developer news in a TV frame for an always-on office screen.
Team analytics for Claude Code and Codex sessions, built from uploaded transcripts.
Cloud security scanner with an AI layer that reads findings without write access to your cloud.
Ruby client that lets agents call tools directly from a JSON manual instead of an MCP server.
Android SSH client whose Rust helper keeps your shell running on the server.
Cache-aware model router that picks a cheaper model for each coding-agent turn.
Agent and LLM tracing stored next to your logs, metrics and traces.
Issue-to-PR pipeline run by coding agents, with human gates between stages.
Open-source CRM whose AI agents have to ask before they change records.