Developer Tools and Model Serving #60

June 27, 2026 — 2 min read

Today's Letter

Vercel added Web Analytics querying to the Vercel CLI through the `vercel metrics` command
The command can return page views, visitors, and custom event data for Vercel projects
The feature is positioned for traffic analysis, trend comparison, and site performance questions from the command line
Vercel says coding agents can use the CLI access to answer questions such as weekly traffic gains, monthly UTM signup performance, and mobile-versus-desktop conversion comparisons
Supported metrics, dimensions, filters, and query options are documented separately in the Vercel docs
The change was published in the Vercel changelog on 2026-06-26
Cross-source confirmation is limited, but the feature is explicitly described on Vercel's official site

Source: vercel.com

GitHub released GitHub Desktop 3.6.0 on June 26, 2026 for macOS and Windows.
The update adds Git worktree support for working on multiple branches in parallel without extra clones.
GitHub positions worktrees as a better fit for coding-agent workflows that open isolated parallel sessions.
Copilot features in GitHub Desktop now run on the Copilot SDK, a shared foundation for AI-assisted workflows.
Commit message generation now reads .github/copilot-instructions.md and AGENTS.md.
Generated commits also follow repository commit metadata rules for style and standards alignment.
Merge conflict handling now includes Copilot-assisted explanations and suggested resolutions before merge completion.
Every Copilot feature now includes a model picker, and BYOK can connect third-party or local models.
GitHub Desktop remains free, while Copilot-powered features require GitHub Copilot access.

Source: github.blog

Hugging Face published a workflow to launch a private OpenAI-compatible vLLM endpoint on HF Jobs with a single `hf jobs run` command.
The post uses `vllm/vllm-openai:latest`, an `a10g-large` GPU flavor, exposed port `8000`, and `vllm serve Qwen/Qwen3-4B` as the baseline example.
Prerequisites are `huggingface_hub>=1.20.0`, local `hf auth login`, and an active payment method or positive prepaid credit balance.
The endpoint is reachable through the HF Jobs proxy and accepts OpenAI API-style requests from curl or the OpenAI Python client.
Access is gated, not public; each request must include an HF token with read access to the job namespace.
Billing is usage-based, and the post lists `a10g-large` at `$1.50/hour`; Hugging Face recommends cancelling the job explicitly when testing is done.
The same pattern can scale to larger models, including `Qwen/Qwen3.5-122B-A10B` on `h200x2`, with `--tensor-parallel-size 2` for multi-GPU serving.

Source: huggingface.co

Jocoletter curates AI, software, and product trends for developers and builders.

#GitHub #HuggingFace #Vercel