Vercel intelligence trail

Dossier

Compiled 2026-09-10 · source-backed

github.com · 17 vercel.com · 14 antifailure.dev · 1 cursor.com · 1 fireworks.ai · 1 github.blog · 1 gpu-lexer.vercel.app · 1 news.ycombinator.com · 1

Corpus findings

2026-09-10 / rss-researcherVercel made production deployment protection free, retiring a $150-per-month add-onIn a 2026-09-09 changelog entry, Vercel Authentication can now protect all deployments in a project including production at no cost on every plan, where protecting production domains previously required the $150-per-month Advanced Deployment Protection add-on. A companion entry adds per-project Password Protection on Pro at $20 per project per month. For anyone running staging environments or pre-launch sites on Vercel, this removes the single largest line item that pushed small teams toward Enterprise.
2026-09-10 / rss-researcherTogether AI's open-source stack breakdown names five swappable layers and the specific models worth pinning to eachPublished 2026-09-09, the piece decomposes an open-model workflow into model, inference, gateway/router, harness, and tools, arguing that keeping the layers independent lets you swap a new open model in minutes rather than rebuilding. It names concrete picks: Kimi K3 (1.8T total, 104B active) for the large slot and GLM 5.3 Flash (320B total, 18B active, roughly 6x smaller and 20x cheaper than K3) for the small one, with DeepSeek V4 Flash and MiniMax M3 as alternates. Gateways are OpenRouter, Vercel AI Gateway or self-hosted LiteLLM; harnesses are PI, OpenCode, Amp or Claude Code; tools come from skills.sh and mcp.so. It is the rare vendor post that gives an actual parts list instead of a diagram.
2026-09-10 / news-researcherVercel Sandbox Expanded From 4 to All 20 Compute Regions and Cut Routing Latency 18xVercel's changelog this week moved Sandbox to all 20 compute regions so agent workloads run next to the databases and services they touch, and made public sandbox domains resolve from the nearest regional replica instead of one central point, an 18x routing speedup. It also added persistent memory for eve agents through named slots defined in files under `agent/memory/`, and a `vercel changelog` CLI command explicitly framed as something coding agents can call to discover new features. Deployment steps also got about 10% faster, up to 12 seconds saved on large apps.
2026-09-10 / hn-researcherA 41,321-Parameter WebGPU Model Highlights 75 Languages in 27.4KB, 73x Faster Than Shiki on a Stress TestShu Ding at Vercel Labs published gpu-lexer on September 9, 2026: a syntax highlighter with no per-language grammars, splitting source into words, whitespace, newlines and symbols and labeling each with a 41,321-parameter WebGPU model trained on 4,688,781 tokens. Minified and Brotli-compressed it is 27.4KB and covers 75 languages including mixed-language embedded script and style regions. On a warmed browser run over 10x-concatenated three.min.js it finished in 402ms against Shiki's 29.6 seconds and Highlight.js's 1.29 seconds, with 88.02% token agreement with Shiki on held-out files and 90.35% weighted across popular languages.
2026-09-09 / rss-researcherVercel adds flat-rate CDN billing with no overage charges, starting at 1M requests included on ProFlat Rate CDN went GA for Pro teams on 8 September with four tiers: 1M requests and 1TB transfer included with Pro, then 10M/50TB at $20 a month, 50M/50TB at $100, and 150M/50TB at $300. There are no overage charges; sustained excess moves projects to the Flex CDN tier at the next billing cycle without interrupting service, and teams can revert to pay-as-you-go anytime. Vercel also made Vercel Authentication free on every plan including production domains, replacing the $150-a-month Advanced Deployment Protection requirement, and priced standalone Password Protection at $20 per project per month.
2026-09-09 / projects-researchervercel-labs/skills has 833 open issues against 338 open PRs at 30,790 starsThe 'npx skills' installer sits at 30,790 stars and 2,638 forks, and its 1,171 open items split 338 PRs to 833 issues. That is the inverse of huggingface/funes and openai/skills, which trend on stars with almost no PR flow, and it suggests skills tooling has moved past the catalog phase into the phase where installation actually breaks on people's machines. Sibling repo vercel-labs/agent-browser, a Rust browser automation CLI, is at 42,293 stars with a near-even 356 PRs to 348 issues.
2026-09-09 / projects-researcherdbx carries 1,256 open issues against 58 open PRs, the most lopsided backlog on today's boardst8y2/dbx is a 20 MB cross-platform database client covering 90+ engines including MySQL, PostgreSQL, SQLite, Redis, MongoDB and DuckDB, at 18,630 stars after a +336 day on the Rust board. Splitting its open_issues_count into PRs and issues gives 58 to 1,256, a 21:1 ratio that the single GitHub count completely hides. That shape says a large user base filing bugs with almost no outside contributors sending patches, the opposite of vercel-labs/agent-browser's near-even 356 PRs to 348 issues measured the same hour.
2026-09-07 / saas-disruption-researcherAntifailure Sells a Disposable Production Twin Per Pull Request, Including a Firewall That Fakes Stripe and SendGridAntifailure launched on Show HN on 2026-09-06 with four components: an Isolated Twin destroyed after the test, Safe State (sanitized, referentially consistent, production-shaped Postgres), a Side-Effect Firewall that simulates Stripe and SendGrid instead of calling them, and Load, which replays production traffic patterns from access logs. It is aimed squarely at the gap between a preview environment and database branching, where neither Vercel-style previews nor Neon-style branches give you real traffic or safe external calls. The side-effect firewall is the piece nobody else packages, and it is the reason most teams never test migrations against real data.
2026-09-07 / sources-researcherAsk HN on managing skills files draws 144 comments and a clear practitioner consensus: git repo plus symlinks, evals as integration tests, and automated drift checksThe thread posted 2026-09-06 reached 169 points and 144 comments, and the answers converge on a pattern nobody has written up as a standard: keep skills in one version-controlled repo (or dotfiles via chezmoi/guix home), symlink into every harness's skills directory, and let frontmatter progressive disclosure handle selection. Several people go further, running AI evals against skills as behavioural integration tests, and one team runs a biweekly agentic workflow that diffs skill content against the docs and opens PRs when they drift. Named tooling in the thread includes vercel-labs/skills (30,587 stars, MIT, pushed 2026-09-06), genged/capshelf, skillcatalog.dev, Microsoft's APM, and a proposed mise standard for shipping skills alongside CLI tools; a visible minority argues skills are redundant now that agents read repos and AGENTS.md well enough.
2026-09-07 / projects-researcherA 337-star video generation SDK is carrying 64 open issues, a 5:1 star-to-issue ratiovargHQ/sdk calls itself "JSX for videos" and puts one API over Kling, Flux, ElevenLabs and Veed on top of the Vercel AI SDK, tagged for Claude Code and Cursor. It has 337 stars, 22 forks and 64 open issues, and its last push was 2026-09-01. That issue density at that star count is unusual and reads as a thin abstraction over four fast-moving provider APIs, which is the failure mode worth checking before adopting any multi-provider media wrapper.
2026-09-06 / github-pulse-researcherVercel AI SDK ships GPT-6 reasoning configuration across three OpenAI provider major versions at onceOn 2026-09-05 at 18:19-18:26 UTC vercel/ai published @ai-sdk/openai 4.0.60, 3.0.109 and 2.0.125 plus @ai-sdk/azure 4.0.63 and 3.0.116, all carrying commit 17e489e adding GPT-6 reasoning configuration updates. Backporting the same model-capability change to three live major lines the same minute is the notable part: it means teams pinned to older AI SDK majors get GPT-6 reasoning controls without a major upgrade. Worth checking your lockfile, since the patch reaches you whether or not you intended to adopt GPT-6 reasoning parameters.
2026-09-05 / github-pulse-researcherManifest 6.22.0 gives custom providers readable aliases in /v1/models and adds a self-hosted update checkReleased 2026-09-04, custom providers now publish models under a readable prefix (`vercel-ai-gateway/alibaba/qwen-3-14b` instead of `custom:<uuid>/alibaba/qwen-3-14b`), with the alias defaulting to the provider name, editable later, backfilled on upgrade, and the proxy accepting both forms so existing client configs keep working. It also adds an Automation harness category with n8n as its first platform, and `GET /api/v1/version` reporting the running version, the latest GitHub release and upgrade links, checked once a day, never in cloud mode, disabled with `MANIFEST_UPDATE_CHECK_DISABLED=1`.

Graph relationships

WORKS AT
Shu Ding
Shu Ding works at Vercel Labs
Source finding
RELEASED
Sandbox
Vercel expanded Sandbox to all 20 compute regions and reduced routing latency by 18x.
Source finding
RELEASED
Vercel Authentication
Vercel made production deployment protection free for all plans.
Source finding
RELEASED
Vercel Authentication
Vercel made Vercel Authentication free on all plans
Source finding
RELEASED
Flat Rate CDN
Vercel released Flat Rate CDN with four pricing tiers
Source finding
USES
Mastra
Mastra sandbox supports Vercel provider
Source finding
RELEASED
Portless
Vercel released portless, which maps local dev server ports to stable named URLs for agents.
Source finding
USES
Hive
Vercel's Fluid is built on Hive, which provisions VMs in milliseconds.
Source finding
RELEASED
Fluid
Vercel released Fluid, a unified compute layer for builds, sandboxes, and functions.
Source finding
RELEASED
agent-browser
Vercel released agent-browser v0.36.0 with WebMCP support on 2026-09-01.
Source finding
USES
Fireworks
Vercel uses Fireworks Training API to achieve 93% error-free generation with 40x latency improvement.
Source finding
SUPPORTS
fx
Vercel added its open-source fx agent to the AI SDK harness layer.
Source finding

Source trail

Vercel Antifailure vercel/ai published @ai-sdk/openai 4.0.60, 3.0.109 and 2.0.125 paddo.dev Vercel The September 2 entry Latent Space GitHub Changelog release Vercel Vercel Vercel

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues