Primary agent. Infrastructure, web design, front-end, back-end, TypeScript. Where most of the actual building happens.

Sourcegraph's agent. Used for think-out-loud mode — interview the other agents' work, edit, piece it all together.

OpenAI's terminal agent. Sys admin level tasks, scripts, operational work. Lighter footprint than Claude Code.

AI-native editor. Best for large codebase navigation and multi-file edits.

Session persistence, window splits, detach/attach. Still the baseline.

Dev agent multiplexer for git worktrees. Route multiple Claude Code instances across isolated branches.

Modern terminal workspace. Layouts as YAML, plugin system, batteries included.

Native macOS terminal built for agentic multitasking. Vertical tabs, notification rings when an agent needs attention, in-app browser. Built on libghostty. Free and open source.

AI terminal. Block-based output, command search, team sharing.

Remote VM instances for long-running Claude Code sessions. Overnight tasks with permissionless execution, committed back to repo.

SSH client. Remote into exe.dev instances, set tasks running, pull results back to local when done.

Vault synced across all devices. Agents write project MDs, specs, captain's logs of sessions. Nothing lost to power outages or session compaction.

Mesh networking across devices and VMs. The plumbing that makes the remote setup feel local.

Stream Deck+ controller for Claude Code. Hardware-to-CLI bridge for agentic workflows.

Autonomous experiment loop on Raspberry Pi. Set a question, walk away.

IDE with browser, multi-agent, PR submission/review/merge all in one place.

Voice-to-agent input on macOS. Talk to your agent instead of typing.

Autonomous coding agent in VS Code. Full file and command access.

OpenAI's long-running agent service. Linear → isolated workspace → Codex per issue.

AI-to-production UI. Prompt → deployed React app. Formerly GPT Engineer.

Vercel's AI component generator. shadcn-native output, direct to codebase.

Design, prototype, and dev handoff. Still the standard for collaborative UI work.

Hand-drawn whiteboard diagrams. The system design tool.

Collaborative infinite canvas. SDK available for embedding.

Frontend + serverless. Auto-deploy from git. This site.

Zero-config static hosting from any repo. Push to main, it's live. Free for public repos.

Edge hosting + Workers. Zero cold starts, D1 database.

Code + deploy in browser. Fast for prototypes and sharing live demos.

Deploy any directory in seconds. `vercel --prod` from any project root.

Instant terminal deployments. Minimal config, immediate URL.

`surge` CLI. Static sites live in under 10 seconds. No config.

Built-in and free. Shift+Cmd+5 opens the recording panel — pick screen or window, set mic, trim on save. Shift+Cmd+3 full screenshot, Shift+Cmd+4 region, Shift+Cmd+6 Touch Bar.

Free, open-source screen recording for demos. No watermarks. Alternative to Screen Studio.

Best capture tool on macOS. Screenshot, record, annotate, share.

Fastest inference available. LPU hardware, near-instant responses. Free tier with rate limits.

Unified API across every major model. One key, any model. Many models available free with rate limits.

Gemini 3.5 Flash leads on agentic benchmarks at fraction-of-frontier cost. Gemini 3 Ultra for max long-context work. Free tier via AI Studio.

Alibaba's open model family. Strong at code and multilingual tasks. Run locally for free via Ollama or LM Studio.

Real-time web search grounding without Google dependency. 2,000 free queries/month.

Desktop app for running local models. Best UX for trying models offline. Good model browser.

Distributed inference across a cluster of consumer devices. Chain Mac Studios over Thunderbolt 5 to run 70B–400B+ models that don't fit on any single machine. Auto-discovers peers via mDNS. MLX backend on Apple Silicon.

CLI-first local model runner. `ollama run llama3`. Best for scripting and agent pipelines.

Open-source desktop alternative to LM Studio. Fully offline, privacy-first.