Tools

Apache-licensed desktop, web and daemon code brings agent sessions, plans and reviews together. Self-hosting still needs integration keys and deployment work.

A local Claude Code mod checks subscription usage, parallel agents and heavy conversations before more work starts. Version 1.1 adds a near-limit wind-down reminder.

A free, MIT-licensed Divvy-style window manager with keyboard presets, multi-monitor placement, and an overlay that now shows already-snapped windows.

A newly packaged Rust web framework combines server-rendered HTML, explicit Markdown representations, and small interactive islands; client-side Rust remains a restricted subset.

A declarative terminal UI library with a browser playground; its October 7 release adds wrapping for styled text. Local integration still needs a host.

A local code-relationship graph adds HTML/CSS impact queries, with CLI and MCP access; source-available licensing and unresolved runtime behavior remain important limits.

A local Rust tool combines code relationships, documents, and past agent sessions into cited context; setup also changes agent settings and Git hooks.

An MIT-licensed patch analyzer and browser leaderboard for code growth, with a newly corrected Go benchmark release—not a maintainability score.

Free cross-platform speech-to-text with local recognition, optional AI cleanup, and real Linux desktop caveats.

Manim-style scenes, local video export, and storyboard checks for agent-written animations—with compatibility and licensing caveats.

A cross-platform skill manager adds collection installation and batch updates while preserving each skill's tool targets and project scope.

Open-source parametric CAD with desktop and browser builds, a headless command interface, and agent tools for modelling, measurement and export—with explicit geometry limitations.

A local proxy and browser dashboard for prompt size, cache reuse, and CPU spillover—not a general-purpose cloud LLM monitor.

A free front end for Hermes, OpenClaw and ACP coding agents, with important differences in how sessions and scheduled jobs work.

Free voice typing with Whisper, Parakeet or Apple speech, custom vocabulary and optional meeting notes—plus privacy settings worth checking.

Track App Store and Google Play keywords, compare market-specific listings, and let an agent query your history without handing it write access by default.

Preference Model opens a Python framework that separates agent work from trusted grading, with VM-backed isolation and submission checks.

An MIT-licensed sitemap crawler, retrieval pipeline, and embeddable chat widget; small-site economics are plausible, but private-doc access control is not included.

Connect to a coding-agent pane from iPhone or iPad, get optional blocked/done alerts, review diffs, and answer in a real terminal.

Timed and untimed commitments in one ordered list, with iCloud sync and an MCP connector that grants agents read and write access.

A self-hosted execution service with per-request isolation requirements, credential-brokered API access, and checked run records; early-stage and not independently audited.

A macOS menu-bar monitor projects when Claude Code, Codex and other coding subscriptions will hit their limits, rather than just showing usage so far.

A public-beta endpoint for classifying text and images with probabilities, fixed choices, or rubric scores; one model and input-only billing.

The open-source cross-device agent adds explicit model choices for chat, screen operation, images and clips, without silently charging its community relay when your own key fails.

A desktop app lets an agent pause for structured human approval of code comments, plans, lists, or images, then continue with the decision.

A Rust/GTK desktop client for Gmail, Microsoft and standard mail accounts; local-first sync, calendars and an opt-in assistant, with important packaging and privacy trade-offs.

An MIT-licensed Chrome and Firefox extension for page and document summaries, with on-device inference or loopback Ollama and llama.cpp connections.

Run pi sessions in separate containers on your own Linux host, with a CLI and shared control plane; useful isolation, but not one VM per task.

A new beta search endpoint for agents via AI Gateway, with Ceramic, Exa, and Linkup as provider choices; useful plumbing, not a new Cloudflare search index.

A new Python package bundles small task-specific classifiers for spam, intent and topic routing; the encoder is larger than the classifier, and this is not zero-shot classification.

A hosted shared workspace lets Codex, Claude Code and other MCP-capable agents exchange notes and files across machines without committing work-in-progress to a repo.

An open-source sidecar for persistent agent-made cards, pages and interactive artifacts, with a local server and a browser or Mac viewer.

An output style and per-prompt reminder for Claude Code, with an optional context meter; useful if the default prose gets in the way, but its savings are developer-reported.

A local Rust graph database with Datalog queries and both recorded-at and effective-at history; the latest release improves query performance and durability, but has a known multi-value bug.