Valet Cloud
New features, improvements, and fixes to Valet Cloud, each week.
See everything your services do, and who did what.
Organization drains. Send logs, traces, and metrics to any OpenTelemetry vendor, for the whole organization or one service.
Audit log. Valet records every write, allowed or refused. Admins can review it.
Service tokens for private apps. Let scripts, cron jobs, and CI call a private app. Mint one with valet apps token mint.
One-step rollback. Ask your agent to roll a site, app, or agent back to an earlier release.
Bigger apps and databases. Run processes and Valet PostgreSQL on economy-1x to economy-8x, or standard sizes that never sleep.
Secrets stay out of chat. You enter secrets on a secure dashboard page, never in the conversation.
Better starting points. Agents build new apps and scheduled agents from working starters.
Background workers now stop cleanly when an app deploys.
Control who can change what, and talk to any agent in the browser.
Agent URLs. Every agent now has its own chat page.
Developer and Admin roles. Give each member a role. Admins manage people, billing, and secrets.
Invitations tab. See and revoke pending invitations on the Team page.
Slow-starting apps get up to 110 seconds to wake before a request times out.
Before a sharing change, the CLI names who gains access and which connected accounts they reach.
Dashboard sessions show who started each one.
A sign-in loop caused by a stale cookie.
Removed members now lose every path back into the organization.
Trigger apps from webhooks and cron, and keep them awake when it matters.
Webhooks and schedules for apps. Trigger an app from a webhook or a cron schedule.
Always-on apps. standard-1x never idles. economy-1x sleeps when quiet.
Visitors see a starting page while an app wakes, and private apps start waking during sign-in.
Faster rebuilds for Go and Python apps. Build logs stream immediately.
App logs show where each start's time goes, so you can fix a slow wake.
Agents connected to Valet's MCP server find and use your organization's skills by name.
Attaching a database and setting variables takes one restart, not two.
valet deploy exits non-zero when a newer deploy replaces it. valet orgs destroy asks first.
Share know-how across your organization, and build per-person apps.
Skills as services. Each skill has its own URL, access, and sharing, and appears in the Skills gallery.
Visitor identity. Apps and sites receive a verified header naming each visitor. No sign-in code needed.
Custom checkers. Plug your own release check into a policy by URL.
Share apps by email. Share an app with specific people, the same way you share sites and skills.
Follow a build. valet builds show -f streams an app's build log.
Release policies now appear in the dashboard.
New pricing. Pay for the sites, apps, and agents you run. No seats, so invite your whole team.
Check every publish for risky code, and manage skills like any other project.
Release policies. Run checks on every publish before it goes live. Each one can block a release or only report.
Vulnerability and secret scans. Built-in checkers find dependencies with known vulnerabilities and credentials left in source.
Skills in Git. A skill can hold many files, and every publish is a version.
Skills page. Browse your organization's skills from the dashboard and the launcher.
Publish with Git. Your agent publishes with a Git push that sends only what changed, so large projects deploy faster.
Your agent can change a few files of a skill without republishing all of it.
Re-sending a site share keeps the access terms you set.
Set your organization's standards once, and keep apps behind Valet sign-in.
Governance. A Governance page shows the skills your organization expects every agent to follow.
Built-in standards. Default security, dependencies, and platform skills teach agents to build safely on Valet.
App access. Apps are private by default and take the same private, password, and public modes as sites.
Models for apps. Attach the valet-inference resource to call a supported model from your app.
Launcher. Your organization's home puts its sites, apps, and agents in one place.
Errors now name what failed and what to do next, with a code you can look up.
Slack agents now know their real bot handle.
Run Procfile apps on Valet, each with its own PostgreSQL database.
Apps. Deploy a Procfile app to its own URL. Valet builds it, runs it, and sleeps it when idle.
Valet PostgreSQL. Give an app a database with valet resources create.
Publish apps from your agent. Your agent can create, publish, rename, and delete apps, not only sites.
Apps page. See every app in the dashboard, whether it's running, and the database it holds.
New models. Claude Sonnet 5.5, GPT-6 Luna, and GPT-6.1 Sol are available to agents.
Zero-downtime deploys. An app keeps serving while its new release starts. Builds and restarts are faster.
Signed in to Valet? Private and password-protected sites let you straight in.
valet sessions transcripts show each tool's name and output.
Newer models, one more connector, and a smoother dashboard.
GPT-5.6 and Claude Opus 5. Choose GPT-5.6 Sol, Terra, or Luna, or Claude Opus 5. Older models are retired.
Porkbun connector. Manage domains and DNS from an agent.
Refresh your organization's design system when your website's style changes.
Integrations you already added stay in the picker, marked Added.
A long-open site tab no longer lands on an expired sign-in page.
The PostHog connector connects again.
Pages pick up your company's style, and one service can back several connectors.
Your brand, automatically. Valet derives your organization's design system from its company website.
Multiple connections. Connect the same service twice, such as two calendars, each with its own key.
GLM-5.3. Agents can run on GLM-5.3 and GLM-5.3 Flash, with image input.
AgentKey connector. Add AgentKey from the connector catalog.
Connector errors now say what the upstream service refused.
The X connector lists who liked or reposted a post.
Destroy a site or an agent with one click.
Logging out now ends your open site sessions too.
Approve with a reaction, and pick up where a conversation left off.
Slack reactions. A reaction can wake an agent, so people can approve with a click.
Metabase and Granola. Add Metabase from the connector catalog and Granola as a webhook channel.
Conversations first. Opening an agent lands you in a conversation, with past ones beside it.
Messages sent while an agent is working wait their turn.
Every supported model now uses its full context window.
Profile pictures appear across the dashboard.
A deploy that races another change now merges instead of failing.
Add skills, catalog connectors, and custom MCP servers from one dialog.
Add integration. Add a catalog skill, write your own, or connect a custom MCP server from one dialog.
Agree connector. Add Agree from the connector catalog.
A rebuilt Sites page.
The Team page has a Settings tab.
Publish from any MCP client, share by email, and keep pages live.
MCP server. Connect https://api.valet.dev/mcp to publish and manage sites from your agent.
Share by email. Share a private site with one person, and revoke it at any time.
Design systems. Pages follow your organization's design system. Choose a preset or bring your own.
Live sites. A site can call the connectors attached to it, so pages show current data.
Name a site when you publish it over MCP.
Rebuilt Agents and Integrations pages, and a one-page first-run setup.
An organization can now hold up to 1,000 sites.
Editing a Slack message no longer wakes an agent in a DM.