Four free tools, four completely different catches

The Lens

SGLang has become one of the default engines for serving open-weight LLMs in production at scale. It runs from a single GPU up to large clusters, with prefix caching, speculative decoding, and day-zero support for new model releases. The project says it runs on hundreds of thousands of GPUs, at xAI, NVIDIA and the major clouds among others. Apache-2.0, free. Install is a uv pip install, with builds for NVIDIA, AMD, Intel, Google TPUs and Ascend NPUs. Models span Llama, Qwen, DeepSeek, Kimi, GLM, Mistral, diffusion models, and most of Hugging Face. It is OpenAI-API compatible, so existing clients drop in without rewrites, and it doubles as a rollout backend for reinforcement learning training. Solo developers and small teams running open-weight models: this is one of the strongest options on the shelf, especially on DeepSeek or with heavy agentic workloads. Large teams running production inference at scale: you are very likely already evaluating it. The catch is that serious inference is still serious work. Cold starts, KV cache tuning, and multi-node setups need real engineering. SGLang gives you a faster engine, not a managed platform. RadixArk, a venture-funded company among its maintainers, is building one on top, with no public pricing yet.

The Lens

Recordly is the open source take on Screen Studio. Record your screen, get auto-zooms on your cursor, polish the timeline, drop in a webcam overlay, and export to MP4 or GIF. The output looks like product marketing video, not a raw screencap. Built with Electron and TypeScript, with native capture where it counts: ScreenCaptureKit on macOS, Windows Graphics Capture plus WASAPI audio on Windows. That matters because the lazy alternative is browser-style screen APIs that drop frames and look amateur. Linux is the weak spot, it records through Electron's own capture APIs and cannot hide the cursor. AGPL 3.0, with prebuilt releases for all three platforms and an AUR package for Arch. Solo devs making demo videos, tutorial creators, and small teams shipping product walkthroughs get the biggest win here. Screen Studio, the closest commercial equivalent, is macOS only and now runs on a subscription: $20/mo billed monthly, or $9/mo billed yearly. Recordly is free on all three platforms, and the editor covers most of what you'd actually use. The catch: AGPL is the strong copyleft license, and Recordly layers two more terms on top of it. You cannot use the Recordly name or branding for your own project, and if you build on the code you have to credit Recordly in your user-facing UI. Add the AGPL rule that offering a modified version as a service means publishing your changes, and anything you intend to ship needs a read of the license first. For recording your own demos, none of it applies.

The Lens

commerce-agents is Anthropic's reference blueprint for building two things on Claude: a shopping agent you embed for customers, and a merchant agent your staff use to run the back office. It is not a product you install, it is example code (real runnable agents across retail, travel, telecom, and entertainment) that shows how to wire the pieces together. Apache-2.0. The value is the pattern, not the demo. Each agent is defined once (prompt, skills, tool contracts, approval gates) and runs on the Messages API, the Agent SDK, and Managed Agents. There is a Claude Code plugin that scaffolds a new agent against your own systems. Nothing here places a real order or charges a card; every write is staged until a person approves it. If you are building commerce agents on Claude, start here instead of from a blank file. It shows the safety scaffolding (the fencing, the staged writes, the approval surface) that you would otherwise learn the hard way. The catch: the code is free but Claude is not. You bring an API key, and cost scales with usage. This is a starting point wired to Anthropic's own stack, not a vendor-neutral framework.

The Lens

OpenWhispr is voice-to-text for your desktop. Hit a hotkey, talk, and it types into whatever app you're in, and it'll transcribe meetings with speaker labels too. The transcription runs on local models (Whisper or NVIDIA's Parakeet) with no API keys required, so your audio never has to leave your machine. MIT, free, cross-platform. Install is trivial, there are prebuilt installers for Mac, Windows, and Linux. Local models download on first use, and if you've got a GPU it'll use Metal, CUDA, or Vulkan to speed things up, but CPU works fine. You can optionally bring your own cloud LLM keys for cleanup, but you don't need to. Everything you need for personal dictation is free and local. There's a separate hosted OpenWhispr Cloud for teams: free up to 2,000 words a week, $6.67/user/mo for Pro, $16.67/user/mo for Business with speaker labels and agent mode. Solo users should just run the free desktop app and skip the cloud entirely. Teams that want shared, unlimited transcription and sync are who the paid tiers are for. This is a real alternative to Otter and Rev if you value keeping audio local. The catch is the usual local-versus-cloud tradeoff. The free local app is good, but the polished team features, unlimited cloud transcription, sync, speaker labels at scale, sit behind the subscription, and the free cloud tier's 2,000-words-a-week cap is tight. For one person it's all upside. For a team, price the Business tier against what you're paying Otter now.