Skip to main content

Console overview

The console at https://console.codyhill.dev is the web home of the Crusoe Agent Platform. This page gives you the full tour: how to sign in, how to pick a project, what every page does, and the keyboard shortcuts that make it fast.

What the console is

The console is a single web app for the whole platform. Anything you can do with the platformctl CLI or the raw API, you can do (or at least see) here: deploy and chat with agents, deploy functions, run serverless services, manage MemoryStore instances, VectorDB indexes, Pub/Sub topics, secrets, service accounts, and MCP servers, and review quotas and the audit log.

A 30-second mental model

Your browser never holds your access token. It holds a session cookie, and a small server behind the console swaps that cookie for a token on every request it forwards to the platform APIs.

Two things follow from this design:

  • The console makes no permission decisions itself. Every "can you do this?" check happens in the service API behind it. What you see in the console is exactly what your account can do.
  • The cookie is HttpOnly, which means JavaScript running in the page can never read your credential. Even a script-injection bug could not steal it.

Sign in

Open https://console.codyhill.dev and enter your email and password.

There is no self-service sign-up page. Accounts are created by an administrator or through an invitation link — ask your administrator for one. See Create an account.

A few things to know about sessions:

  • If your account was created with a temporary password, the console shows a sticky "Choose a new password" dialog before anything else works. New passwords must be at least 12 characters.
  • A session lasts up to 12 hours. The cookie does not survive closing your browser — you sign in again next time. This is deliberate.
  • Signing out asks the server to clear the session cookie and returns you to the sign-in screen. If the server is unreachable the screen still resets, but the cookie is not cleared — and because session tokens are stateless, a token already issued stays valid until its 12-hour expiry either way.
  • The console never stores your password.

The scope switcher and command palette

Everything in the console is filtered by your scope: the organization and project you are working in. A project is a private workspace — every resource you create belongs to exactly one.

Press Cmd+K (Mac) or Ctrl+K (Windows/Linux), or click the Scope button in the top bar. One box does two jobs:

  • Search. Type a few letters and it finds anything you can see across all your projects: agents, functions, serverless services, MCP servers, service accounts, secrets, and projects. Each result is a deep link.
  • Switch project. Under a Switch project heading, one row per project you belong to — pick one and your scope changes everywhere.

Use the arrow keys to move, Enter to open, Esc to close. The console remembers your last scope, so you land back where you were.

What page does what

The left navigation is organized by section. Pages you cannot use are hidden, and pages you can see but not fully use render a plain-language "no access" explanation instead of a raw error.

SectionPageWhat it doesWho sees it
OverviewHomeDashboard tiles counting every resource type in the projectProject members
IntegrationsCrusoe CloudLinks this project to Crusoe Cloud resourcesProject members
ComputeServerlessDeploy and manage scale-to-zero services, traffic splits, triggersProject members
ComputeFunctionsDeploy one-file HTTP handlers; invoke and read logsProject members
ComputeAgentsDeploy agents, chat with them, inspect sessions, manage the embed widgetProject members
ComputeMCP serversPublish tool code as hosted MCP endpoints, with versions and rollbackProject members
Data servicesMemoryStoreCreate and manage key-value store instancesProject members
Data servicesVectorDBCreate indexes, browse data, run live queriesProject members
MessagingPub/SubTopics and subscriptionsProject members
SecuritySecretsStore secret values, manage versions and bindingsProject members
SecurityService accountsMachine identities and their API keysProject members
SecurityMy API keysYour personal API keys (not tied to any project)Every signed-in user
SecurityAudit logWho changed what, newest firstAnyone who can see the project
ProjectIAM & membersMembers, roles, invitations, break-glass grantsAnyone who can see the project
ProjectQuotasLive capacity bars for the project's limitsProject members
AdministrationOrganizationsOrgs and the platform user directoryPlatform admins
AdministrationProjectsThe projects you belong to; create new onesEvery signed-in user

Clicking a row usually opens a detail page. For example, an agent's detail page has tabs for Overview, Test (chat with it), Logs, Revisions, YAML, Access, and Embed. The active tab is part of the URL, so you can share a link straight to it.

Platform admins see less, not more

A platform admin can see every project's IAM page but not its contents — no agents, no secrets, no data. Service pages explain this and point to break-glass access. Read why platform admins cannot read your data.

The docs drawer

Every service page has a Docs button in its header. It opens a reference drawer for that service — field meanings, limits, and API routes — without leaving the page. The content is built into the console, so it works even with no internet route to this site.

Live updates

List and detail pages refresh themselves: roughly every 5 seconds while something is building or deploying, and about every 25 seconds once things are steady. Polling pauses entirely while the tab is hidden. The Refresh button on any page bumps it back to the fast cadence.

Deploying from the browser

The Agents and Functions pages can deploy code straight from the browser, in three modes: write (edit files in the page), files (pick or drag a folder), or archive (upload a ready-made .tar.gz). The browser packs your files into a compressed archive and uploads it.

Limits when the browser does the packing: 1 MiB per file, 16 MiB total, and file paths up to 100 bytes. A supplied archive can be up to 32 MiB. Junk directories like .git, node_modules, and __pycache__ are skipped automatically. Two sharp edges:

  • An upload replaces the whole stored source tree for that agent or function.
  • The write and files modes store source as text — binary files must go through the archive mode.

If your archive is larger than 32 MiB, the console refuses with this exact message and you should use the CLI, which uploads directly to the agent engine (its own cap is 100 MiB):

the uploaded source archive is larger than the console accepts (32 MiB); deploy it with `platformctl deploy` instead, which uploads to the agent engine directly

When something goes wrong

Every console error carries a request id. Error messages are deliberately scrubbed of internal details, so the request id is the string to quote to support — it lets them find the exact log lines across the console and the service behind it.

How it compares

Crusoe Agent Platform consoleThe big clouds
Scope of one screenOne console for agents, functions, data services, messaging, and IAMGCP Console, AWS Management Console, and Azure Portal cover far more services (VMs, networking, billing, hundreds of products)
Sign-inEmail + password; accounts by admin or invitation only; no MFA yet (alpha)All three offer MFA, SSO, and federated identity
Staff access to your dataStructurally blocked; only a visible, audited break-glass grant opens itGoverned by internal provider controls, not visible to you in-product
SearchCmd/Ctrl+K palette across every resource you can seeComparable global search in all three

Next steps