DeepSeek V4.1 Flash 1017 Ultrafast model added to AI Gateway, delivering up to 1,000 tokens per second and a 1 M‑token context window for fast interactive coding and agentic workflows.
Observability changelog digest
Nov 2 - Nov 8, 2026. Useful releases, risky migrations, and noisy updates from the Observability channel.
No matching updates in this bucket.
Observability updates in 2026-w45
Adds DeepSeek Search and Fetch tools to AI Gateway for page search and markdown retrieval via a single API key
Chat SDK agents can now stream generative UI by emitting a JSON spec that json-render turns into native cards (e.g., Slack Block Kit, Teams Adaptive Cards).
Added --delete-project-after flag to `vercel deploy` to set an expiration for newly created projects (supports 1h, 1d, 1w, 2w, 30d).
Vercel Tracing is generally available on all plans with dashboard and CLI search, KQL filtering, and live mode.
Token Pass credit is automatically applied to AI Gateway usage for any auth method, removing the need for a dedicated Token Pass API key.
Introduces a dedicated PostHog Laravel package with lifecycle-aware context for requests, queues, commands, exceptions, and feature flags.
AI Gateway now displays usage activity for stored provider keys, including masked previews, aliases, and last‑used/updated dates.
Added support for SHA-384 hash pinning of the core SDK script.
Added support for sending transactional emails from PostHog templates in Python and Node.js SDKs.
Ampersand (humansand/ampersand-v0.1) is now available on AI Gateway, free through Nov 13 2026, for group‑chat scenarios with send_message and pass function tools.
Adds autocapture of service metrics for Python services using OpenTelemetry, covering HTTP requests, database operations, and runtime health.
Add cross‑project dashboards allowing insights from multiple projects on a single view
Admins can review public dashboard links when removing a member.
Adds integration to connect Seer Agent with a Grafana account.
Vercel Schedules is now in public beta on all plans, enabling recurring or one‑time tasks in production, preview, or custom environments.
Data Warehouse tables can now mask selected source columns using deterministic hashes before storage.
Added option to hide specific node types in Data Warehouse lineage view.
Endpoints with unsupported variable expressions now fail clearly instead of silently retrying.
Gemini Omni 1.2 Flash added to AI Gateway, enabling text‑to‑video generation (3‑15 s clips, up to 30 s total) with native audio.
Workers AI will deprecate numerous LLM and other models on November 15, 2026; update your apps to remove references.
Redesigned API to use Go's context, removing hubs
Data Warehouse person-property syncs now resume from the last confirmed chunk after interruption.
Agents can verify the integration source for a Data Warehouse destination directly from the original connect link.
Added support for materializing endpoints that use non‑aggregate conditional functions like `nullIf` and `multiIf`.
Adds Self-Driving inbox prioritization that ranks reports based on activity, priority, freshness, and user needs.
Setup wizard now prevents webhook-synced tables from proceeding without a configured webhook
Data Warehouse syncs now safely hand off during empty API pages instead of timing out
Added ability to pin Data Warehouse views to a slower refresh cadence, showing downstream views that will read older data.
14 data warehouse source integrations (e.g., Help Scout, Easybill, Steam, Cloudflare D1, etc.) have graduated to GA and are production‑ready.
Runtime automatically retries fetch() and RPC calls to Durable Objects on retryable errors, up to 4 attempts within 10 seconds.
Opening AI chat from an inbox report now automatically includes the report's evidence, artifacts, and linked tasks.
Added atomic start() that admits a run and claims a hook in one step, eliminating the getHookByToken‑then‑start race.
Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed are now available on AI Gateway from multiple Zero Data Retention (ZDR) providers, expanding beyond the previous Moonshot AI‑only support.
Vercel now scans public sources (GitHub, GitLab, Docker, npm, PyPI, Hugging Face, Android APKs, Firefox Add-ons) for exposed tokens via partnership with Truffle Security.
Introduce model aliases to represent a pool of models under a single name
All major server SDKs (Java, .NET, Elixir, Go, Ruby/Rails, PHP, Rust, Node.js) now propagate trace IDs and request context into error events and application spans.
Add experimental `mcp` feature to Rust SDK for sending tool-call data to MCP analytics.
Autocapture can now append the value of input, textarea, or select fields when those fields are explicitly opted‑in.
Vercel Security Review enters public beta for Pro and Enterprise, automatically scanning every push to connected GitHub repositories for vulnerabilities.
Vercel CI can now run on a private GitHub repository without needing a linked Vercel project.
Introduces the `vercel doctor` command to diagnose Vercel CLI setup problems.
New AI Gateways require authentication by default and cannot be disabled via dashboard or API.
Agents SDK v0.33.0 upgrades to Pi 1.5, cutting storage writes (~27× for long answers) and memory use while boosting throughput (up to 459 runs/sec in benchmark).
Tencent Hy4 model added to AI Gateway with reasoning, tool calling, structured outputs, image input, 1M‑token context, vision and computer use capabilities.
Crawl results now include an "outcome" field for each URL (e.g., success, robots_disallowed, timeout, page_error).
Added regional request distribution view on the R2 bucket Metrics page.
Vercel adds support for a vercel.toml configuration file, using TOML format with comments and a flatter structure.
Vercel MCP now accepts OAuth clients that publish a Client ID Metadata Document (CIMD) and use its HTTPS URL as the client_id
Vercel MCP now shows deployments as an interactive card in chat clients that support MCP Apps, displaying live status, build logs, source/commit details, and a Visit button.
Redesigned Flags Explorer UI now groups overrides, branch recommendations, and actions together.
Vercel Sandboxes that mount Drives start up to 2.4× faster (median 0.7s vs 1.7s, p95 1.6s vs 2.2s).
Adds Project Connections so apps can evaluate flags from another Vercel project in the same team without using SDK keys, authenticating via deployment identity.
Adds `defaults: 'latest'` to automatically adopt defaults from future browser SDK upgrades.
Flagship bindings can now evaluate flags against a local store during development, enabling offline testing.
AI Search billing start date postponed to Dec 1 2026; no charges before then and pricing remains unchanged.
Add support for initializing the web SDK with a single external script tag that queues calls until the full client loads
Vercel scans lockfiles and blocks known malicious npm package versions before and during dependency installation.
LongCat 2.5 from Meituan is now available on AI Gateway, supporting text and image inputs with a 1M-token context window for code, documents, and extended task histories.
JavaScript Workers with compatibility date ≥ 2026‑11‑17 will use the new module registry by default, removing the need for the `new_module_registry` flag.
Workers with a compatibility date of 2026-11-17 or later receive client disconnect events via request.signal by default, removing the need for the enable_request_signal flag.