Telexed

AI news through a solo-operator lens — only what changes your day22 of 829

FILTER[All][Agents & tools][Models & API][Generative media][Infra & SaaS][ASO & growth][Indie business][Idea signals][Other][★6+ high-signal]

vercel_blog ✕clear filters

Fri, May 293 dispatches

#0829
#0829Infra & SaaS vercel_blogyesterday
`Vercel` switches function invocations to per-unit billing
100radar
VercelFrontend cloud platform — optimized for Next.js deployment
Function usage now maps directly to actual calls. The rate is unchanged, but billing becomes smoother for small teams and bursty apps.
- Pro customers pay $0.0000006 per invocation, equivalent to the previous $0.60 per 1M invocations package rate.
- The change starts from the next billing cycle; current effective rates stay unchanged until the existing cycle ends.
- Per-unit billing reduces the chance that Vercel Functions usage eats a large chunk of monthly included credits at once.
- Applies to Pro and new Enterprise customers, so active Vercel apps should recheck cost dashboards after renewal.
Source: vercel.com/changelog/function-invocations-now-billed-perRead original →
FIG-8291:1
100radar
FIG-8291:1
#0828
#0828Infra & SaaS vercel_blog2 days ago
Vercel warns AI endpoints are prime targets for inference theft
70radar
BotIDVercel bot defense tool — per-request automation checks
Cheap HTTP checks do not match $2-per-prompt economics. Verify every AI request, or one abused endpoint can turn into a five-figure bill.
- Attackers wrap custom endpoints in OpenAI or Anthropic adapters, then resell stolen inference through standard SDK clients.
- Session auth and IP limits fail when proxy fleets and throwaway accounts spread calls across thousands of identities.
- Playgrounds are the highest-risk shape because users control prompts, models, and parameters. Fixed server prompts lower but do not remove risk.
- BotID is Vercel's proposed gate: run bot verification on every AI request, not just signup or session start.
Source: vercel.com/blog/protecting-against-inference-theftRead original →
FIG-8281:1
70radar
FIG-8281:1
#0827
#0827Infra & SaaS vercel_blog2 days ago
`Vercel Sandbox` now runs Docker containers
80radar
Vercel SandboxCloud sandbox — isolates agent-driven development work
Agents can now build images, run services, and install system packages inside isolated sandboxes. This turns preview and test environments into fuller dev machines.
- Docker can be installed with a daemon inside Vercel Sandbox, so agents can run containerized apps without touching the host system.
- Redis, Postgres, and other test dependencies can run as containers; fewer external services are needed for isolated QA.
- With persistent sandboxes, installed Docker and pulled images carry across sessions, cutting repeated setup time.
- FUSE filesystem drivers and VPN clients are now supported too, widening what sandboxed agent workflows can access.
Source: vercel.com/changelog/run-docker-containers-inside-vercelRead original →
FIG-8271:1
80radar
FIG-8271:1

Thu, May 283 dispatches

#0826
#0826Models & API vercel_blog2 days ago
`Claude Opus 4.8` lands on `Vercel AI Gateway`
80radar
Vercel AI GatewayLLM gateway — unified calls, cost tracking, and failover
Long-running coding and knowledge tasks can now route through one Vercel API. Useful if you already want retries, cost tracking, and failover without wiring providers yourself.
- Use anthropic/claude-opus-4.8 in the AI SDK; model switching stays inside the existing Gateway path.
- Positioned for long-horizon agent work: multi-step refactors and complex coding tasks need less mid-task correction.
- Gateway adds usage and cost tracking, retries, failover, and latency/cost-based provider sorting. Less infra glue around LLM calls.
- Pricing mirrors provider rates with no Vercel markup or platform inference fee, including BYOK requests.
Source: vercel.com/changelog/opus-4-8-on-ai-gatewayRead original →
FIG-8261:1
80radar
FIG-8261:1
#0825
#0825Infra & SaaS vercel_blog3 days ago
`Amazon OpenSearch Serverless` lands in the `Vercel Marketplace`
70radar
Amazon OpenSearch ServerlessServerless search service — unified vector and keyword search
Search infra can now be provisioned from the Vercel dashboard. Built-in env injection, scale-to-zero, and AWS credits make it easier to test production-style search without ops overhead.
- Provisioning happens inside Vercel; env vars are injected automatically, cutting setup friction for search-backed apps.
- One collection supports vector, lexical, hybrid, and agentic search, so RAG and normal site search can share infra.
- Claims include 20x faster autoscaling, scale-to-zero, and up to 60% cost savings versus peak-capacity pricing.
- New AWS accounts created through Vercel get $100 credits; existing AWS accounts can be linked and monitored from the dashboard.
Source: vercel.com/changelog/amazon-opensearch-serverless-is-nowRead original →
FIG-8251:1
70radar
FIG-8251:1
#0824
#0824Infra & SaaS vercel_blog3 days ago
`Vercel AI Gateway` Adds Team-Wide Provider Allowlist
60radar
AI GatewayAI routing SaaS — one API across multiple model providers
Provider routing can now be locked at the account layer. Useful when client, compliance, or cost rules require approved AI vendors only.
- The allowlist applies to every AI Gateway request, including BYOK traffic. Request-level filters cannot bypass it.
- Filtering is by provider, not model. If one provider fails, Gateway can fall back only to another allowed provider for that model.
- New providers are disabled by default once the allowlist is on, so vendor scope will not expand silently.
- Only team owners can change the list. Coding agents are also blocked from routing to unapproved providers.
Source: vercel.com/changelog/team-wide-provider-allowlist-on-ai-Read original →
FIG-8241:1
60radar
FIG-8241:1

Wed, May 271 dispatches

#0823
#0823Infra & SaaS vercel_blog4 days ago
`Vercel CLI` Adds Experimental Native Binaries Without Node.js
60radar
Vercel CLIDeployment CLI — manages and deploys Vercel projects
Deploy commands can now run from signed native binaries with no Node runtime. Faster startup and OS-scoped credential storage make it worth testing on daily deploy machines.
- Install via @vercel/vc-native; -f is required because it uses the same global vercel and vc bin names as the standard CLI.
- Binaries are code-signed, so the OS can verify they came from Vercel and were not modified.
- On macOS, credentials move into the system Keychain scoped to the binary, reducing token exposure to other local processes.
- Support covers macOS, Linux, and Windows on x64 and arm64; useful for lean CI images and fresh machines.
Source: vercel.com/changelog/experimental-native-binaries-for-veRead original →
FIG-8231:1
60radar
FIG-8231:1

Tue, May 262 dispatches

#0822
#0822Infra & SaaS vercel_blog5 days ago
`Vercel Sandboxes` Persistence Is Now GA
80radar
Vercel SandboxesCloud execution sandbox — automatic filesystem restore
Filesystem state now survives sessions by default. Long-running agent and preview workflows get simpler, but snapshot storage becomes a new cost line.
- Sandbox.create() now enables persistence by default, so files are saved and restored without manual snapshot handling.
- Each sandbox gets a customizable durable name. Sandbox.getOrCreate() adds idempotent retrieve-or-create for long-lived environments.
- Automatic snapshots consume separately billed snapshot storage. Use --non-persistent for throwaway jobs to avoid idle storage cost.
- Calling runCommand() or writeFiles() on a stopped sandbox resumes from the latest snapshot, which fits agent task handoff and preview reuse.
Source: vercel.com/changelog/sandbox-persistence-is-now-gaRead original →
FIG-8221:1
80radar
FIG-8221:1
#0821
#0821Infra & SaaS vercel_blog5 days ago
`Firecrawl` joins the `Vercel Marketplace`
60radar
FirecrawlWeb crawling API — extracts LLM-ready structured data
Teams can add crawling infrastructure directly from the hosting stack. It turns web pages into LLM-ready data, useful for RAG and agent features without running scrapers yourself.
- Firecrawl can scrape pages into markdown, HTML, structured data, or screenshots, cutting the glue code around ingestion.
- Search plus full-page retrieval works in one call, a practical fit for small RAG tools and monitoring agents.
- Dynamic page interaction via AI prompts or code reduces the pain of JavaScript-heavy sites, but usage cost still needs checking.
Source: vercel.com/changelog/firecrawl-joins-the-vercel-marketplRead original →
FIG-8211:1
60radar
FIG-8211:1

Thu, May 213 dispatches

#0820
#0820Infra & SaaS vercel_bloglast week
`Vercel CLI` Can Now Pull Anomaly Alert Details
60radar
Anomaly status now reaches the terminal, including AI investigation output with --ai. Useful for incident triage, but gated behind Observability Plus.
- vercel alerts lists team or project alerts with start time, alert type, and active status; terminal triage gets less context-switching.
- --ai adds AI investigation results next to each alert, giving coding agents enough context to act from the CLI.
- Availability is limited to Observability Plus, so the value depends on whether observability spend is already justified.
Source: vercel.com/changelog/pull-anomaly-alert-details-using-thRead original →
FIG-8201:1
60radar
FIG-8201:1
#0819
#0819Infra & SaaS vercel_bloglast week
`Qwen 3.7 Max` lands on `Vercel AI Gateway`
70radar
Qwen 3.7 MaxLLM — built for coding and long-horizon agents
Alibaba’s agent-focused model is now callable through the AI SDK. Routing, retries, cost tracking, and BYOK make it worth a quick prototype test.
- Use model ID alibaba/qwen-3.7-max; existing AI SDK paths can compare it without provider-specific wiring.
- Positioned for coding, office workflow automation, and long-horizon autonomous execution. Add it to multi-file engineering bake-offs.
- Gateway features include usage and cost tracking, retries, failover, and intelligent routing. One API reduces provider-switching overhead.
- Built-in BYOK, observability, and custom reporting keep model experiments easier to measure across small products.
Source: vercel.com/changelog/qwen-3-7-max-now-available-on-verceRead original →
FIG-8191:1
70radar
FIG-8191:1
#0818
#0818Infra & SaaS vercel_bloglast week
Configure weighted traffic splits for Vercel Flags from the Vercel CLI
60radar
The new vercel flags split command allows for weighted traffic distribution for feature flags. This simplifies setting up simple A/B tests without leaving your terminal, useful for quick experiments.
- You can run the command interactively or pass flags for automated environments.
- This removes the need to switch to the Vercel dashboard for managing traffic splits, keeping you in the development flow.
- It's designed for splitting traffic between two variants, making it ideal for gradual rollouts or simple UI changes.
Source: vercel.com/changelog/configure-weighted-traffic-splits-fRead original →
FIG-8181:1
60radar
FIG-8181:1

Wed, May 205 dispatches

#0817
#0817Models & API vercel_bloglast week
`Grok Build 0.1` lands on `Vercel AI Gateway`
60radar
Grok Build 0.1Agentic coding model — powers the Grok Build CLI
A beta agentic coding model is now callable through the AI SDK. Useful for quick experiments, but early access and fixed reasoning keep it from being a default choice.
- Set the model to xai/grok-build-0.1 in the AI SDK; integration cost is low if the app already uses AI Gateway.
- Reasoning effort is not configurable, and there is no non-reasoning mode. Latency and cost controls are limited for production agents.
- AI Gateway adds usage tracking, cost reporting, retries, failover, BYOK, and provider routing around the model call.
Source: vercel.com/changelog/grok-build-0-1-now-available-on-verRead original →
FIG-8171:1
60radar
FIG-8171:1
#0816
#0816Infra & SaaS vercel_blog2 weeks ago
`Vercel Chat SDK` Adds `callbackUrl` for Card Buttons and Modals
60radar
Vercel Chat SDKChat UI SDK — integrates Vercel workflows and adapters
Agent runs can now pause and resume from UI actions. Useful for approval flows in Slack or Teams without extra glue code.
- callbackUrl sends button-click or form-submit payloads to your endpoint, so workflows can resume from external UI events.
- Cards need a workflow webhook URL wired into each button’s callbackUrl; modals include submitted form data in the payload.
- Button callbacks work on most official-adapter platforms, while modal callbacks are limited to Slack and Teams for now.
Source: vercel.com/changelog/chat-sdk-now-supports-callback-urlsRead original →
FIG-8161:1
60radar
FIG-8161:1
#0815
#0815Infra & SaaS vercel_blog2 weeks ago
`Chat SDK` adds message subjects and direct adapter SDK access
60radar
Chat SDKBot SDK — adapters for GitHub, Linear, and Slack
Bots can now resolve the parent GitHub issue, PR, or Linear item from a mention. This makes workflow bots less glue-heavy and worth adopting for support or triage automation.
- message.subject returns title, status, URL, and typed parent payload for GitHub and Linear comments; Slack returns null.
- Subject lookup is cached per message, so repeated reads avoid extra provider API calls in the same handling path.
- Adapters now expose underlying platform SDKs directly; custom provider API calls no longer need separate client wiring.
Source: vercel.com/changelog/chat-sdk-adds-message-subjects-and-Read original →
FIG-8151:1
60radar
FIG-8151:1
#0814
#0814Infra & SaaS vercel_blog2 weeks ago
`Chat SDK` adds built-in `AI SDK` tools via `chat/ai`
70radar
Chat SDKVercel chat app SDK — built-in AI tool wiring
Agent wiring is now one call: createChatTools(chat). Approval-gated writes and preset-scoped tools reduce custom glue code for Vercel chat apps.
- createChatTools(chat) connects Chat SDK read/write actions into an agent, cutting boilerplate around chat state operations.
- Write tools default to requireApproval, so destructive agent actions start behind an approval gate.
- reader, messenger, and moderator presets construct only allowed tools, keeping tool exposure narrower by default.
Source: vercel.com/changelog/chat-sdk-now-includes-ai-sdk-toolsRead original →
FIG-8141:1
70radar
FIG-8141:1
#0813
#0813Infra & SaaS vercel_blog2 weeks ago
`Vercel AI Gateway plugin` connects WordPress to 40+ AI providers
80radar
Vercel AI Gateway pluginWordPress AI connector — one-key routing to 40+ providers
WordPress sites can now route AI features through one gateway key instead of per-provider setup. Useful for shipping AI plugins with less credential, fallback, and billing plumbing.
- The connector works with the new WordPress AI Client, which requires WordPress 7.0; older sites need an upgrade first.
- One AI Gateway key in Settings > Connectors replaces separate credentials for Anthropic, Google, OpenAI, xAI, DeepSeek, and more.
- Plugins built on WordPress AI Client inherit the connector automatically, reducing custom provider integration work.
- Supported modes include text, structured JSON, image generation/editing, and video through the same prompt builder.
- Automatic fallbacks and dynamic model discovery lower outage handling and model-update maintenance.
Source: vercel.com/changelog/vercel-ai-gateway-plugin-for-wordprRead original →
FIG-8131:1
80radar
FIG-8131:1

Tue, May 195 dispatches

#0812
#0812Agents & tools vercel_blog2 weeks ago
`Nuxt MCP Toolkit` adds support for MCP apps
70radar
Nuxt MCP ToolkitMCP toolkit for Nuxt — Vue SFC-based tool UIs
Agent tools can now return inline interactive HTML, not just text. Useful for richer tool flows inside Claude or ChatGPT.
- Tools declared with defineMcpApp can render interactive HTML responses in MCP clients such as Claude and ChatGPT.
- useMcpApp lets the UI read pre-hydrated data, trigger follow-up prompts, or call other tools from inside the response.
- Vue SFCs are bundled into self-contained HTML at build time and served from the MCP endpoint, reducing custom UI plumbing.
Source: vercel.com/changelog/nuxt-mcp-toolkit-mcp-appsRead original →
FIG-8121:1
70radar
FIG-8121:1
#0811
#0811Models & API vercel_blog2 weeks ago
`Gemini 3.5 Flash` lands on `Vercel AI Gateway`
80radar
Vercel AI GatewayLLM API gateway — routing, cost tracking, and failover
Better coding, reasoning, and agent loops are now available through one gateway call. Good fit when uptime, cost tracking, and failover matter more than direct provider integration.
- Use google/gemini-3.5-flash in the AI SDK; migration is mostly a model string change for existing Gateway users.
- Default thinking level is medium, aiming for faster and cheaper generation without fully dropping reasoning quality.
- temperature, topP, topK, and thinking_budget are unsupported, so apps relying on sampling controls need compatibility checks.
- AI Gateway adds usage tracking, cost reporting, retries, failover, BYOK, and provider routing on top of the model call.
Source: vercel.com/changelog/gemini-3-5-flash-on-ai-gatewayRead original →
FIG-8111:1
80radar
FIG-8111:1
#0810
#0810Infra & SaaS vercel_blog2 weeks ago
`Vercel` opens `Flat Rate CDN` limited beta for Pro teams
100radar
Flat Rate CDNVercel CDN plan — fixed monthly billing regardless of traffic
CDN bills can now be capped with a fixed monthly fee. This reduces launch-risk from viral traffic, bot spikes, and route mistakes.
- Flat Rate CDN replaces usage-based CDN billing with a fixed monthly price for Pro teams in limited beta.
- Edge Requests, Fast Data Transfer, and related Vercel CDN products are no longer metered under the plan.
- Temporary traffic spikes will not raise the bill, which makes launches and experiments easier to budget before the month starts.
Source: vercel.com/changelog/flat-rate-cdn-in-limited-betaRead original →
FIG-8101:1
100radar
FIG-8101:1
#0809
#0809Infra & SaaS vercel_blog2 weeks ago
`Vercel Sandbox` Now Runs `Claude Managed Agents`
80radar
Vercel SandboxIsolated runtime — per-session Firecracker microVMs
Agent tool calls can run inside isolated Vercel microVMs while secrets stay outside the sandbox. Useful for production agent features that need private APIs and tighter network control.
- Claude Managed Agents provides the model, harness, tools, and session state; Vercel supplies the execution environment for self-hosted tool calls.
- Each session gets its own Firecracker microVM with millisecond startup, making per-user agent isolation practical.
- Credential brokering happens at the firewall, so secrets do not enter the sandbox. This is the right default for customer-data workflows.
- Deny-by-default egress with a domain allowlist reduces blast radius when agents call tools or browse controlled services.
Source: vercel.com/changelog/run-claude-managed-agents-with-vercRead original →
FIG-8091:1
80radar
FIG-8091:1
#0808
#0808Infra & SaaS vercel_blog2 weeks ago
`Vercel` stops charging for WAF-mitigated traffic
90radar
VercelFrontend cloud platform — integrated Next.js deploys and CDN
Blocked bots no longer turn into surprise CDN bills. This directly lowers the risk of scrapers, credential stuffing, and abused API endpoints on small apps.
- Traffic denied, challenged, or rate-limited by Vercel Firewall now waives CDN Requests and Fast Data Transfer charges.
- DDoS mitigation was already unlimited and free; this change covers non-DDoS abuse handled by custom rules, managed rules, and rate limits.
- The waiver applies automatically to every project using Vercel Firewall; no config change or migration required.
- Product pages, login routes, and expensive API endpoints are the obvious places to add stricter rules now.
Source: vercel.com/changelog/web-application-firewall-mitigated-Read original →
FIG-8081:1
90radar
FIG-8081:1

`Vercel` switches function invocations to per-unit billing

Vercel warns AI endpoints are prime targets for inference theft

`Vercel Sandbox` now runs Docker containers

`Claude Opus 4.8` lands on `Vercel AI Gateway`

`Amazon OpenSearch Serverless` lands in the `Vercel Marketplace`

`Vercel AI Gateway` Adds Team-Wide Provider Allowlist

`Vercel CLI` Adds Experimental Native Binaries Without Node.js

`Vercel Sandboxes` Persistence Is Now GA

`Firecrawl` joins the `Vercel Marketplace`

`Vercel CLI` Can Now Pull Anomaly Alert Details

`Qwen 3.7 Max` lands on `Vercel AI Gateway`

Configure weighted traffic splits for Vercel Flags from the Vercel CLI

`Grok Build 0.1` lands on `Vercel AI Gateway`

`Vercel Chat SDK` Adds `callbackUrl` for Card Buttons and Modals

`Chat SDK` adds message subjects and direct adapter SDK access

`Chat SDK` adds built-in `AI SDK` tools via `chat/ai`

`Vercel AI Gateway plugin` connects WordPress to 40+ AI providers

`Nuxt MCP Toolkit` adds support for MCP apps

`Gemini 3.5 Flash` lands on `Vercel AI Gateway`

`Vercel` opens `Flat Rate CDN` limited beta for Pro teams

`Vercel Sandbox` Now Runs `Claude Managed Agents`

`Vercel` stops charging for WAF-mitigated traffic