Changelog
What is new in Vallkey
Features, improvements and fixes, as they ship.
Latest
vallkey connect sets your project up
Vallkey CLI 0.9 installs the package and changes how your app starts, asking first. Python 0.4 turns on by itself, and hooks 0.2 name each app.
- Newvallkey connect installs the package and edits what really starts your app: the start script, every Procfile process, the final command of a Dockerfile, railway.json, railway.toml and render.yaml. It asks before each change. With --yes it now really changes files and installs packages; --print only shows the steps.
- Newvallkey verify with no arguments waits for your app's requests and shows each one as it arrives, so no request ID needs copying. vallkey init offers to run connect straight away.
- ImprovedPython 0.4: installing the package turns auto-connect on, with no variable and no import, and processes that send no requests pay no startup cost. On AWS Lambda, make import vallkey.auto the first line of your handler; vallkey connect adds it in SAM and Serverless Framework projects.
- ImprovedHooks 0.2 for Ruby, Java, .NET, PHP and Erlang name the app on requests already addressed to Vallkey, so one app shows as one place. Moving an app to a hook no longer sets off an alert that it stopped.
- Securityvallkey doctor and connect previews read a project's environment folder instead of running its Python, so they stay read-only in a cloned project.
Deleted keys keep their history
A deleted key stays readable, with what it did, for your plan's activity days. Then its history is erased.
- NewDeleted keys are listed under Deleted keys, read-only: their requests, blocked requests, places, usage and speed, who deleted them and when. Their real key is destroyed at once, as before.
- NewThe account's Activity and its exports keep deleted keys' requests, marked deleted, and ended agent tasks stay in the Agent tasks list.
- ImprovedRequests made with a key deleted this month still count toward this month's usage.
- SecurityA deleted key's history is erased after your plan's activity days (7, 30 or 90), and at once with the account when the account is deleted.
Verify your app's connection
Vallkey CLI 0.8 adds setup guidance, local diagnostics and verification of an exact application request, now tested on Vercel.
- Newvallkey connect recommends address settings or hook instructions for your project. It previews setup without editing files. vallkey doctor checks local readiness and distinguishes it from a verified deployment.
- Newvallkey verify checks a recent x-guard-request-id from your application's response. It verifies successful requests or expected rule denials, without making provider calls.
- ImprovedNode/Bun hooks handle malformed queries and more body credential formats. Recognized protected requests do not silently fall back after routing errors. Gateway settings are validated and debug logs omit paths and queries.
- ImprovedSetup guidance preserves existing Next.js instrumentation and explains native hook installation. Support depends on the HTTP transport and credential format; verify your deployed app.
- ImprovedTested on a real Vercel deployment (Next.js 16, Node.js 24): fetch and node:https requests with a ghost key go through Vallkey, rule denials come back as expected, other requests are left alone, and vallkey verify confirms each one. Edge routes still need the API address setting. AWS Lambda has not been tested yet.
- ImprovedTested on Railway with Node.js, Python, Ruby, Java and .NET. Railway, Render and Heroku also use a service's variables while building it, so NODE_OPTIONS or RUBYOPT there would break the build: the hook now goes in the start command. Java and .NET use a Dockerfile, or on Railway without Docker a railpack.json that sets the variable for the running app only. vallkey hook, the dashboard and the docs say so.
- ImprovedTested on Vercel with Next.js, Node.js, Bun, Python and Ruby functions. Vercel's variables also reach the build, so for Node.js, Bun and Ruby functions the hook is now the first line of each function; Python only needs VALLKEY_AUTO=1.
- ImprovedCLI 0.8.1: vallkey hook writes instrumentation.js in a JavaScript Next.js project, and vallkey doctor reads the instrumentation file instead of always warning about it.
Set up a key without waiting
Fix what your app needs during a key's first day, apply waiting changes with your passkey, and protect a key in as many places as you like.
- NewSetup window: during a key's first day, looser rules and revivals apply as soon as you save them. Your team is still alerted about every change.
- NewApply now: after the first day, the account owner can let waiting looser rules or a replacement real key take effect at once, after confirming with a passkey.
- ImprovedPlans count real keys, not ghost keys. Add another place for a key you already protect as often as you need, on every plan.
- ImprovedWider default rules. Stripe keys allow subscriptions, setup intents, saved payment methods, the billing portal, and capturing or cancelling payments. OpenAI keys allow moderation and fetching stored responses. AI keys start at 300 requests a minute.
- ImprovedAI budgets size requests instead of refusing them: images, files, stored conversations and server-side tools are counted, models without a known price are estimated at the highest known list price, and tool definitions are no longer mistaken for media.
- ImprovedAnswers are no longer cut short to fit a budget that has room. When the remaining allowance does lower a request's output limit, the answer says so in x-vallkey-output-limit.
No code changes, in eight languages
Startup hooks route supported HTTP clients through Vallkey in Node.js, Bun, Python, Ruby, Java, .NET, PHP and Elixir.
- Newnpx vallkey hook tells your project's language from its files, downloads the hook where the language needs one, checks its checksum, and prints the one setting to add, with Dockerfile lines when you have a Dockerfile.
- NewHooks for Ruby, Java (and Kotlin, Scala), .NET, PHP and Elixir, next to Node.js and Python. Requests that carry a ghost key go to Vallkey; every other request is left alone.
- NewWhere it's used: each key lists every place it is used from, live, such as Vercel, AWS Lambda or a laptop, with today's requests and what is wrong at each one. npx vallkey status shows the same, and a steady place that goes quiet sends an alert.
- NewThe Connect tab asks where your app runs, and shows exactly what to set there: Vercel, Railway, Render, Fly.io, Heroku, AWS Lambda, Cloud Run, Azure, Kubernetes, Docker or your own server.
- ImprovedS3 uploads that name the bucket in the address, as every AWS SDK does by default, and chunk-signed uploads, the AWS .NET SDK's default, now work through Vallkey.
- ImprovedTested with each language's official SDKs, used as their docs show, in containers cut off from the internet: a request the hook missed fails instead of reaching the provider.
Lost a ghost key? See it again
Admins can see a ghost key again on its key's page, without rotating it and redeploying every app.
- NewYour ghost key, under Connect: confirm it's you and it shows for 30 seconds, with a copy button. Each view is in the audit log and your team gets an alert.
- SecurityOnly admins, only in the dashboard: never through the API, the CLI or AI agents. A ghost key found in public is never shown again.
- SecurityThe owner can turn it off in Settings, Security. Every stored copy is destroyed, and ghost keys are shown only once, as before.
One key, any size
Spread one ghost key over several real keys, let one key take thousands of requests a second, see where the time goes, and manage keys with Terraform.
- NewKey pools: add real keys from other accounts or projects to a ghost key. Requests take turns, a busy key rests while the others carry on, and one the provider refuses is taken out. Team and Business.
- NewHigh-volume keys: every Vallkey server checks the key's requests itself, so one key can take thousands a second. Kills reach every server within a second or two, and money limits stay exact. Business.
- NewSpeed and errors on each key: typical and slow answer times, how long Vallkey's own checks take, what failed at the provider and what the rules blocked, and the slowest endpoints.
- NewOAuth 1.0a signing, with ready-made setups for X, Tumblr, Flickr and the Garmin Health API. Your app can keep its OAuth library and use the ghost key as the consumer key.
- NewA Terraform provider for keys, rules and custom APIs in code, on its way to the Terraform Registry. Rules changed in the dashboard show up on the next plan.
- ImprovedBusiness accounts get priority support, with the answer times from the uptime promise.
Built for busy apps
Fewer provider calls, fewer "too many requests" errors, and a warning before any limit refuses a request.
- NewAnswer caching: answers to the GET endpoints you choose are kept for a few seconds and repeats never reach the provider. Every request is still checked, and each key keeps its own answers.
- NewTrying again: when a provider says too many requests, Vallkey waits as long as it asks and sends the request again, only where that is safe.
- NewPacing: send at most so many requests a second to a provider, and hold the rest for a moment instead of refusing them.
- NewEarly warnings when a provider starts rate-limiting a key, and when one key gets very busy, with what to do about it.
- ImprovedActivity shows answers from the cache, tries and waits, and responses say so in x-vallkey headers.
170 more services, ready to use
Pick DeepSeek, Supabase, Shopify, Slack, Notion, Sentry and more when you create a key: the address, how the key is sent and starter rules are already set.
- NewReady-made setups for the APIs people use most, in AI, databases, email and chat, payments, developer tools, monitoring, work tools, search, sign-in and web3. Each one is checked against the provider's docs.
- NewKeys can send the key in the API's own form, such as Authorization: Token or SSWS, as the Basic auth username, or inside the address path (Telegram, Infura).
- ImprovedThe service picker has filters and search, and keys made from a setup show its name and logo everywhere.
- NewGoogle service accounts for Vertex AI, Firebase Cloud Messaging, Sheets, Drive, Cloud Storage and BigQuery: paste the JSON key file once, and Vallkey handles the access tokens.
- NewS3-compatible storage: Cloudflare R2, Backblaze B2, DigitalOcean Spaces and Wasabi, signed again with the real credentials.
- NewKeys in two parts (Datadog's API and application keys), Zoom's account tokens, and Qdrant Cloud on its own port.
- NewAzure (Resource Manager, Blob Storage, AI Search, Maps), Google Analytics, Compute Engine, Firebase Authentication, WooCommerce, Adobe Commerce, BigCommerce, Ghost, Strapi, WordPress.com, Marketo, ActiveCampaign, OneLogin, X (Twitter), Rollbar, BugSnag, Raygun, Fastly, TomTom, Jenkins and more.
- ImprovedService logos in their brand colours, readable in light and dark mode.
- ImprovedAPIs whose addresses end in a slash, such as Sentry and Supabase, now work with custom keys.
Keys for websites and mobile apps
Call APIs straight from a web page or an app, without putting a key in it that anyone could reuse.
- NewShort-lived keys: your server makes one for each signed-in user. It expires in minutes, can only do what its ghost key allows, and ends when the ghost key is killed or rotated.
- NewAllowed websites: list the websites whose pages may use a key. Pages on other websites are refused.
- SecurityLimits for each visitor and for each user, including a daily AI budget per user, so one person cannot use up a key for everyone.
- NewBlock one user in a click from their request in Activity. Their keys stop at once; nobody else is affected.
- ImprovedActivity shows the website and the user of each request, and can be filtered by both.
- NewcreateSessionKey in the vallkey npm package and create_session_key in the Python package, plus a step-by-step guide on each key's Connect tab.
Confirm it's you, and emails that look the part
Dangerous actions now ask for a quick check even when you are signed in, and every email is clearer.
- SecurityDeleting a key, changing its real key, adding a way to sign in, inviting someone and other dangerous actions ask for a passkey, an authenticator code or an email code. One check covers the next 10 minutes.
- SecurityEmails with codes say when, where and from which browser they were asked for.
- ImprovedEvery email has a clear layout, a large code and a button where there is something to do.
- ImprovedWhen a session ends, you come back to the page you were on after signing in.
- ImprovedA fuller sidebar with your account, help and today's numbers, and a new menu on phones.
Developer tools
Protect a whole project in one command, route any SDK through Vallkey without code changes, and let AI agents and CI help.
- Newnpx vallkey: sign in from the browser, swap the keys in your .env files for ghost keys, and scan code and git history for keys.
- Newvallkey/auto for Node.js: every SDK goes through Vallkey with no base URL to set.
- NewThe vallkey Python package: the same for httpx, requests, aiohttp and urllib, with VALLKEY_AUTO=1.
- NewAn MCP server for Claude Code, Cursor and other AI agents. They protect keys without ever seeing one.
- NewA GitHub Action that fails the build when a key is committed, and marks it on its line.
Plans for every stage
A Free plan to start, Pro and Team for production, and Business for companies with security reviews.
- NewFree, Pro, Team and Business plans, monthly or yearly with 20% off.
- NewBilling in the dashboard: usage this month, plan changes, invoices and your card.
- ImprovedBlocked requests never count toward your plan, and paid accounts are never cut off.
Stop a leaked key before you notice
Keys now learn where they are used, warn you about anything new, and can kill themselves when they look stolen.
- SecurityLimit a key to your servers' IP addresses or to the countries you choose.
- SecurityAutomatic kill on a burst of blocked requests, a new country or a traffic spike.
- SecurityGhost keys and tokens pushed to public GitHub repositories are turned off at once.
- NewSecurity signals on the overview and each key: new countries, new networks, spikes.
- NewRotate a ghost key with a grace period, and see when your app stopped using the old one.
- NewAPI tokens for scripts and CI, each with read, admin or full access.
- ImprovedEvery request shows its IP address, country, client and provider time.
Sign in your way, and a guided start
GitHub and email sign-in, a short onboarding for new accounts, and a redesigned settings page.
- NewContinue with GitHub to sign up or sign in, and connect GitHub from Settings.
- NewSign in with a one-time code sent to your email.
- NewOnboarding for new accounts: your name, your stack and your first key.
- ImprovedSettings, rebuilt with a section menu and clear sign-in methods.
- ImprovedSign-up is now two steps: your email and a code.
A new dashboard
A faster, calmer dashboard with keyboard shortcuts, charts and request details.
- NewCommand menu and keyboard shortcuts for everything.
- NewUsage charts for requests, blocks and AI tokens, over 24 hours to 90 days.
- NewSearch and filter activity, and open any request for its full details.
- NewKill every key at once in an emergency.
- ImprovedHighlighted code and a live JSON editor for rules.
Teams, AWS, PayPal and more
The biggest release yet: team roles, an audit log and five new kinds of protection.
- NewTeam members with owner, admin and viewer roles, invited by link.
- NewAudit log of every change, with CSV export.
- NewAWS request signing, PayPal with OAuth, OpenAI Realtime over WebSockets.
- NewDaily AI token caps and GraphQL operation rules.
- SecurityReplace the real key behind a ghost key, with a one-hour safety delay and alerts.
Any API, and twenty services
Protect keys for any HTTPS API, with built-in support for the most common services.
- NewCustom API keys: choose the address and how the key is sent.
- NewBuilt-in support for 20 services across payments, AI, messaging and developer tools.
- ImprovedLight and dark themes.
Your next leaked key can be a non-event
Protect your first key in five minutes. Free plan, no card needed.