AI radar
AI radar
8 min read
A running log of AI-adjacent links I want to remember: tools to try, projects to watch, articles I read or mean to. Some are tested and I have an opinion. Most are on the want-to-try pile.
Status legend:
- ✅ tried it, recommend
- 🟡 tried it, mixed feelings
- ❌ tried it, would not bother again
- 👀 on my list, not tried yet
| Date added | Item | What it is | Status |
|---|---|---|---|
| 2026-07-24 | Understand Anything | Turns a codebase into an interactive knowledge graph you can click through, search, and ask questions about. Tree-sitter does the deterministic parsing (files, functions, classes, dependencies) and an LLM writes the plain-English summaries on top, so you get domain views, auto-generated architecture tours, layer detection (API, service, data, UI, utility), fuzzy plus semantic search, and change-impact analysis. Ships as a plugin for Claude Code, Codex, Cursor, Copilot, and Gemini CLI, reusing whatever model that platform already has; Ollama for a local-only setup. TypeScript, MIT. | 👀 Not tried yet, looks interesting |
| 2026-07-24 | Orca | “Worktree IDE” for coding agents from Stably: runs many agents in parallel, each in its own git worktree, wrapped in a desktop app with Ghostty-inspired terminals, a file editor, git tracking, inline diff comments, and an embedded Chromium design mode for clicking into UI. 25+ built-in agents (Claude Code, Codex, Gemini, opencode, Copilot, Cursor CLI, Grok, …) using your existing subscriptions. SSH worktrees put agents on a remote box with full editing, git, and terminals plus auto-reconnect. macOS, Windows, Linux, plus iOS and Android apps. MIT. | ✅ Using it, looks OK. Feels a bit heavy on resources, but it is interesting; the remote-server support and the mobile app are what stand out |
| 2026-07-24 | AgentAPI | Go binary from Coder that puts an HTTP API in front of a coding agent’s CLI: agentapi server -- claude runs the real claude binary inside an in-memory terminal emulator and translates API calls into keystrokes. Exposes GET /messages, POST /message, GET /status, an SSE /events stream, an OpenAPI schema at /openapi.json, a built-in chat UI at /chat, and agentapi attach to drop into the live terminal session. Works with Claude Code, Codex, Gemini, Copilot, Amp, Aider, Goose, opencode, Amazon Q, Auggie, and Cursor CLI. MIT. | 👀 Not tried yet, but a good hack to keep in the back pocket: it drives the CLI itself, so subscription auth keeps working even if the SDKs stop accepting it |
| 2026-07-21 | AgentField | Open-source control plane that turns agent code (Python, Go, TypeScript) into callable APIs and microservices. Stateless Go control plane on top: agent mesh with cross-agent calls and tracing, tag-based discovery, schema-constrained output (Pydantic/Zod), async execution with webhooks and SSE, canary deployments, distributed KV plus vector memory, W3C DID identities, and Prometheus metrics with workflow DAGs. Also orchestrates multi-turn coding harnesses (Claude Code, Codex, Gemini CLI, OpenCode). Apache-2.0. | 👀 Not tried yet, looks interesting, will have to give it a spin |
| 2026-07-09 | ccboard | Single Rust binary that monitors Claude Code sessions, costs, config, hooks, agents, and MCP servers. Ships both a 9-tab TUI and a web interface, with live process tracking, budget alerts, and 30-day cost forecasting. MIT. Same category as codeburn. | 🟡 Tried it, so-so UI, uninstalled; may revisit later |
| 2026-06-26 | agent-browser | Fast Playwright-style headless-browser automation CLI built for AI agents: open, read (agent-readable text / accessibility tree), click, type, screenshot, snapshot, eval, CDP connect. Ships version-matched skills (agent-browser skills get core) with copy-paste patterns. Apache-2.0. | ✅ Tested, recommend — I use it to self-verify blog renders against the local Hugo dev server (read the page text, screenshot it) before handing over a preview URL |
| 2026-06-08 | l0-cache | Lightweight Rust CLI proxy that filters and compresses shell-command output before it reaches an AI coding assistant (Claude Code, Gemini CLI): ANSI stripping, line collapsing, diff-context compression, for ~50-80% token savings. Zero runtime deps; macOS/Linux/remote. MIT. Same category as RTK and ccxray. | ❌ Tried it for a few days, didn’t see meaningful token savings in practice, uninstalled |
| 2026-05-27 | sx | Go CLI that acts as a “private npm” for AI assets — skills, MCP configs, slash commands, prompts. Vault-based with scoped install targets (--org, --repo, --path, --team, --user, --bot) so each context only sees the assets it needs. Multi-client (Claude Code, Cursor, Copilot, Gemini, Kiro, OpenCode) and exposes the vault to claude.ai / chatgpt.com via a cloud relay over WebSocket. Apache-2.0. | 👀 Not tried yet |
| 2026-05-22 | cs.sh by jedi4ever | Tiny bash wrapper that runs Claude Code headlessly without -p (to skip the API-mode surcharge): spins up a one-shot session, auto-approves every tool call via a PreToolUse hook, captures the final assistant reply with a Stop hook, prints it on stdout, and cleans up the session file. ~30 lines. | 👀 Not tried yet, probably good |
| 2026-05-22 | tweakcc | CLI tool that patches Claude Code itself to customize system prompts, themes, thinking verbs, spinners, input-box styling, custom toolsets, AGENTS.md support, and unlock private/unreleased features. Works for both native and npm installs. | 👀 Not tried yet, looks cool |
| 2026-05-22 | andrej-karpathy-skills | A single CLAUDE.md file distilling Andrej Karpathy’s observations on LLM coding pitfalls into four principles: think before coding, simplicity first, surgical changes, goal-driven execution. Drop-in for any Claude Code project. | 👀 Not tried yet, looks cool |
| 2026-05-20 | caveman | Claude Code skill that rewrites prompts in “caveman speak” — short, content-word-only phrasing — to cut token usage by ~65%. Half meme, half real prompt-compression experiment; 60k+ stars. | 👀 Not tried yet, looks promising |
| 2026-05-20 | codeburn | Interactive TUI dashboard for AI coding token spend across Claude Code, Codex, and Cursor. Parses local session logs to break down cost per session, model, and tool over time. | 🟡 Tried, mixed feelings — works fine but the data it shows didn’t change how I work |
| 2026-05-17 | Tolaria | Desktop app (macOS/Windows/Linux) for managing Markdown knowledge bases. Files-first — notes stay as plain .md files, never locked in a proprietary format. Git-first — every vault is a git repo with full history. Built by the Refactoring.fm folks; 10k+ stars, used for second brains, personal wikis, and storing AI assistant context. | 👀 Not tried yet |
| 2026-05-13 | claude-account-switcher | macOS menubar app for switching between multiple Claude Code logins. Stores each account’s OAuth tokens in Keychain (claude-switcher:{email}), keeps cached account metadata in ~/.config/claude-switcher/accounts.json, and atomically swaps ~/.claude.json’s oauthAccount field on switch — so other tools (e.g. statusline scripts) can read the active account locally without a network call. | ✅ Tested, recommend |
| 2026-05-12 | dot-ai (my write-up) | Viktor Farcic’s AI-powered MCP server for Kubernetes operations. I use it for shared/git skills — custom prompts synced from a central git repo (my wxs/ai-resources) and exposed as /dot-ai-* slash commands globally across all Claude Code projects. | ✅ Tested, recommend |
| 2026-05-12 | Infisical agent-vault | Open-source credential broker that securely proxies API calls for AI agents by injecting credentials at the network layer — agents never see the secrets directly. | 👀 Not tried yet — waiting on #124 |
| 2026-05-12 | multica.ai | Open-source platform that turns coding agents into real teammates — assign tasks to AI agents, track progress, and manage a combined human-agent workforce from a centralized dashboard. | 👀 Not tried yet |
| 2026-05-12 | clawd-on-desk | Pixel desktop pet that reacts in real-time to AI coding agents (Claude Code, Codex, Cursor, Gemini CLI, …) with 12 animated states. Permission-bubble UI for tool approvals, eye tracking in idle, session dashboard for concurrent agents. | ✅ Tested, recommend |
| 2026-05-12 | agnix | Linter and LSP for AI coding assistants — validates CLAUDE.md, AGENTS.md, SKILL.md, hooks and MCP configs with autofixes. Rust CLI with plugins for VS Code and other major IDEs. | 👀 Not tried yet — waiting on #909 for per-file rule overrides |
| 2026-05-12 | nanoclaw (my deployment: Vasile) | Lightweight container-isolated personal AI agent framework. Connects to WhatsApp, Telegram, Slack, Discord, Gmail; memory, scheduled jobs, runs on Anthropic’s Agents SDK. Built from scratch, ~15 source files. | ✅ Tested, recommend |
| 2026-05-12 | Dippy (my write-up) | CLI tool that runs as a Claude Code PreToolUse hook to manage permissions. Supports wildcards at any position, guidance messages on prompt, layered global + project configs — things settings.json still can’t do natively. | ✅ Tested, recommend |
| 2026-05-12 | RTK (my write-up) | Rust CLI proxy that sits between Claude Code and shell commands, filtering noisy tool output (passing tests, ANSI escapes, repeated log lines) before it reaches the context window. | 🟡 Tried, hit some issues, might revisit |
| 2026-05-12 | ccxray | Transparent HTTP proxy + dashboard that gives “X-ray vision” into Claude Code sessions — inspect every request/response, tool call, and token spend in real time. | 🟡 Interesting, couldn’t find a use case |
Last updated on