Web console¶
One command starts a local, branded console where everything the CLI does is a form or a button:
It opens http://127.0.0.1:7434/ in your browser. The console runs as a user service (launchd on macOS, systemd --user on Linux) so it survives terminal sessions, and it has no CDN, no telemetry and no external calls.
Same commands, same logs
Every click runs the same cloudseed ... command you would type. Its output streams live into the Activity drawer, and it lands in the same audit trail and undo journal. Nothing the console does is invisible to the CLI, or the other way round.
Start, reopen, stop¶
cs enable ui # start the service and open the console
cs enable ui --no-open # start it without opening a browser
cs enable ui --port 7440 # use another port
cs ui # open it again (starts it if needed)
cs ui token # print the link with the access token
cs ui status # enabled? URL, health, service, log
cs ui logs -n 50 # the last lines of the server log
cs disable ui # stop it and remove the user service
Tour¶
The left rail holds ten views; the number keys 1 to 0 switch between them.
| View | What you do there |
|---|---|
| Overview (1) | Counters for environments, clusters, resources in state, running jobs, MCP and agent status; a card per environment with Status, Outputs and Troubleshoot. |
| Create (2) | The environment wizard: pick a cloud, answer the basics, set options, review and run. A live preview shows the exact command it will run. |
| Environments (3) | Each environment's card with Manage: Plan, Change settings, Update my IP, Re-provision, SSH command, Kubernetes, VPN user, Inventory, Cost, Use in terminal, Destroy. |
| Platform (4) | The whole catalog by group, with filters: install, plan or uninstall a group or a single item, read the status from the cluster, expose UIs, list and add nodes, run kubectl and helm. |
| Resilience (5) | Disaster recovery (DR drill, backup now, restore, schedules), chaos engineering (suites, your own workload, stop all), security scans (everything, CIS, NSA / MITRE, vulnerabilities, host CIS, STIG, cloud CIS, FIPS), and Well-Architected assessments with profile and evidence age controls. |
| All actions (6) | Every action as a card with a form: search it, fill it in, run it. The cards say whether an action is read-only or changes things. |
| Reports (7) | The DR drill, chaos and scan reports of the selected environment with their verdicts, plus the logs of its setup, apply and destroy runs. |
| Agents & MCP (8) | Turn agentic mode and the context brief on or off, pick the agent and model, run a task; review usage metadata and download its JSON; deploy the MCP server, connect or disconnect each detected client, self-test, read the connection guide. |
| Credentials (9) | The local credential vault: set, replace or remove cloud keys and API tokens (masked, never shown back). |
| Help (0) | The CLI's help topics and the Explain search. |
The architecture assessment uses the same cs scan architecture engine as the CLI and MCP.
The Agents & MCP usage report needs no environment selection. It shows observed task and MCP activity, with reasons for unavailable metrics, and offers a JSON download. Estimates are not provider bills or remaining quota; the server cannot observe its MCP client's model tokens. See agent usage for the supported measurements. It reads saved configuration and evidence without cloud access or infrastructure changes. Its reports distinguish PASS, FAIL and INCOMPLETE; missing, stale or manual-review evidence never silently passes.
The top bar holds the ⌘K search, the environment selector (the environment the Platform, Resilience and Reports views act on), Undo, the theme switch and the Activity drawer.
Read scan reports¶
Select an environment, open Reports, then choose Open beside a scan. The report shows its verdict, scope, coverage limits and scanner diagnostics. Expand a finding to read its check ID, status, severity, resource, region, observed result, remediation and references whenever the scanner supplies those fields. Nested evidence stays available, including details from Kubernetes, image, host and FIPS scans.
Read Verdict policy alongside the findings: the cloud benchmark fails for any failed observation, while the Kubernetes and image scanners have severity thresholds. A threshold passing does not mean every finding is resolved. Missing evidence for requested checks, execution errors and required manual verification appear as INCOMPLETE, including in Activity. Routine informational entries do not cause that state; disabled or explicitly inapplicable checks show N/A. Open the finding for the specific gap and its next step. Unknown findings include the reason evidence could not establish a result and how to resolve it. Scanner diagnostics also explain report-wide problems such as zero observations, a nonzero exit, access denial or unreachable endpoints; new cloud scans retain redacted diagnostic examples and an output-artifact reference. Older reports may contain only an error count, which cannot identify the original cause without saved diagnostic text. Cloud scans can include resources outside the selected Cloudseed environment; check the report's scope before deciding which resource needs a change.
The compact report view shows at most 100 findings and 200 checks/results, with an explicit notice when more exist. View full JSON and View full Markdown open the complete saved report as text; a scanner's own retention limits and raw-output location remain visible there. Reports above the console's 32 MiB display limit must be opened locally. Older cloud reports display a warning because they may have retained only high/critical failures; rerun the cloud scan to export complete findings and explanations.
An agent launched from the console can review saved results through cs evidence list|read; MCP clients use cloudseed_evidence. Follow every page with the same revision until the artifact is complete. Read the recorded time, diagnostics and coverage limits: unknown=0 does not establish that all resources or controls were checked. See scenario 14's saved-report exercise.
Create an environment¶
The Create view is a four-step wizard built from each cloud's own setup questions, so it always matches the CLI.
%%{init: {"flowchart": {"nodeSpacing": 30, "rankSpacing": 38, "padding": 12}}}%%
flowchart TB
s1["1 Cloud<br/>AWS, GCP, Azure, VMware"] --> s2["2 Basics<br/>name, region, network, state, SSH access"]
s2 --> s3["3 Options<br/>the target's variables:<br/>Kubernetes, VPN, FIPS, sizes"]
s3 --> s4["4 Review and run<br/>Dry run, Plan only or Apply"]- Dry run renders and validates the Terraform, Plan only shows the plan; both are always safe.
- Apply creates billable resources, so it needs the tick "I understand Apply creates billable cloud resources".
- Your answers are kept while you look at other views; Start over clears them.
- Open an existing environment's Change settings and the same wizard edits it instead.
Manage environments¶
On an environment card, Manage opens its actions. Read-only ones (Status, Outputs, Inventory, Cost, Troubleshoot, Plan) run straight away. Anything that changes infrastructure or a host opens a dialog showing the command that will run, and asks for an explicit tick before it starts. Destroy can also delete the remote state storage and the local directory, each with its own checkbox.
Use in terminal makes that environment the CLI's current one (cs env use), so cs kubectl and friends act on it.
Activity, undo and search¶
- Activity (the
`key) streams every job's output live, one tab per job. Closing the drawer does not stop a job. - Undo lists the undo history with what each entry reverts. Undo the newest entry of a scope, or Discard a step that can never succeed. It is the same journal as
cs undo, including global entries (agents, MCP, UI, credentials), which only you can undo. - ⌘K (Ctrl+K) searches actions, environments, catalog items and views, and every explainable name.
"?" explains anything in place¶
A small ? sits beside what the console shows: each view's title, every cloud, field and question of the wizard (with the stack variable it sets), an environment's target, bastion, Kubernetes, VPN and FIPS chips, every platform group and item, the resilience cards, reports, agents, MCP, credentials and every action card and dialog.
Hover or focus it for a one-line summary and the cs explain ... command. Click it to open the Explain panel: the page cs explain prints, laid out as headings, bullets and copyable commands, with a Page | Terminal switch for the exact CLI text. Related pages and "did you mean" suggestions are links (Back returns), and the footer copies the CLI command or opens the page full width in Help. Press ? to explain the view you are on. See Explain everywhere.
Keyboard¶
| Key | Action |
|---|---|
| ⌘K / Ctrl+K | search actions, environments, catalog items, views and explanations |
? | explain the page you are on |
1 ... 0 | switch views |
` | show or hide Activity |
| ⌘B / Ctrl+B | collapse the sidebar |
| Esc | close the explanation, a dialog or the search |
| Alt+← | back, inside the Explain panel |
Security model¶
The console is meant for you, on your machine:
- It listens on 127.0.0.1 only.
- It is token-protected. The token arrives in the link (
cs ui token) and is kept only in that browser tab: no cookie. API calls send it in theX-CS-Tokenheader. - The server checks the
HostandOriginheaders and forbids framing (frame-ancestors 'none'), which blocks DNS rebinding and clickjacking. - It serves only environment logs and reports, never state files or keys.
- Credentials typed into the Credentials view go to the local vault (
~/.cloudseed/credentials.json, mode 0600) and are only ever injected into cloudseed processes. Variables exported in your shell always win. - Destructive actions need an explicit tick, and every action is in the audit trail.
Keep it local
Do not expose the port through a tunnel or reverse proxy. The token grants everything cloudseed can do on your machine, including destroying environments.
Troubleshooting the console¶
| Symptom | Fix |
|---|---|
| "This console link is no longer valid" | the token changed: cs ui token prints the current link |
| The page does not load | cs ui status, then cs ui logs -n 100; cs ui restarts it if needed |
| Port 7434 is taken | cs enable ui --port 7440 |
| An action fails | its output is in Activity and in the environment's log; cs troubleshoot <cloud> --env <name> explains it |
Related: Scenario 15 - web console and FinOps, cs help ui.