Credentials¶
cloudseed works with the logins you already have: AWS profiles and SSO, Google application-default credentials, az login, and the usual environment variables. When you would rather not export secrets in your shell, keep them in cloudseed's local credential vault.
The vault¶
cs creds set AWS_ACCESS_KEY_ID AWS_SECRET_ACCESS_KEY # both are asked with hidden input
cs creds set AWS_PROFILE=prod # KEY=VALUE for non-secret settings
cs creds list # secrets masked, settings shown
cs creds unset AWS_PROFILE
cs creds clear --forget # empty the vault, keep no copy for undo
- Stored in
~/.cloudseed/credentials.jsonwith mode 0600, never printed:listshows secrets as••••••••plus the last four characters of long ones. - Injected as environment variables into every cloudseed command (CLI, web console, MCP server). A variable already exported in your shell always wins, and
listmarks itfrom shell. set KEYwithout a value prompts with hidden input, so the secret never lands in your shell history or the process list. It needs a terminal; without one it stops with exit code 2 and changes nothing.KEY=VALUEis for non-secret settings (profiles, project ids, paths); its value is masked in the audit log anyway. Paths are stored absolute, with~expanded.cs undo --globalrevertsset,unsetandclear: an overwritten value comes back.unsetandclearkeep a copy of what they removed in the undo journal (0600) until five newer global actions push it out;--forgetkeeps none.
What goes in it¶
| Purpose | Variables |
|---|---|
| AWS | AWS_PROFILE, AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_SESSION_TOKEN, AWS_DEFAULT_REGION |
| Google Cloud | GOOGLE_APPLICATION_CREDENTIALS (a key file's path), or GOOGLE_CREDENTIALS (the key's JSON, pasted at the hidden prompt), GOOGLE_PROJECT |
| Azure | ARM_SUBSCRIPTION_ID, ARM_TENANT_ID, ARM_CLIENT_ID, ARM_CLIENT_SECRET |
| Agents | ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, GROK_API_KEY |
| Platform and hosts | UBUNTU_PRO_TOKEN (FIPS), TS_AUTHKEY (Tailscale), GITLAB_RUNNER_TOKEN (GitLab runner) |
| Data platforms | DATABRICKS_TOKEN, SNOWFLAKE_PASSWORD |
Any other variable works too, with one exception: names that change how programs start, which code they load, which files or servers they trust, or where they send requests are refused (PATH, HOME, LD_* / DYLD_*, PYTHON*, ANSIBLE_*, OPENSSL_*, NODE_*, GIT_*, *_CONFIG, *_BASE_URL, *_ENDPOINT*, proxies, TF_LOG*, CLOUDSEED_* and similar). The vault's values reach every cloudseed process, so those belong in your shell if you really need them.
The web console's Credentials view manages the same vault.
Where secrets never go¶
%%{init: {"flowchart": {"nodeSpacing": 30, "rankSpacing": 38, "padding": 12}}}%%
flowchart TB
vault["Vault / shell /<br/>cloud login files"] --> cmd["cloudseed command<br/>(your process)"]
cmd --> tf["Terraform, Ansible,<br/>kubectl, helm"]
cmd -. "redacted output only" .-> agent["AI agent / MCP client"]
cmd -. "never" .-> state["Terraform state,<br/>outputs, remote hosts"]- Not to agents. Agent processes run without credential variables; child cloudseed commands get them back from a per-session socket, and every line of output is redacted. See Agentic mode.
- Not to MCP clients. Tool calls run cloudseed as a child under the same broker, with redacted output; state and credential files are never exposed as resources. See MCP server.
- Not to your hosts. Provisioning copies the repository, never
~/.cloudseed, and leaves out private keys, kubeconfigs,.envfiles,*.tfvars, state and credential files a checkout may hold. A file whose content is a private key stops the copy. - Not to state. The stacks keep no secrets in Terraform state or outputs; the bastion only gets your public key.
- Not to logs. Logs and the audit trail are redacted: AWS keys, private keys, JWTs, API keys and
password=pairs are masked.
Other secret files cloudseed keeps¶
All under ~/.cloudseed, all private to your user:
| File | What |
|---|---|
envs/<cloud>-<env>/ssh/ | the environment's SSH key pair and known_hosts |
envs/<cloud>-<env>/k8s/, vpn/, platform/ | kubeconfig, .ovpn profiles, generated platform passwords (platform/secrets.json, 0600) |
managed.json | Databricks / Snowflake connection profiles |
vmware.json | the vmrest login cloudseed generated |
mcp/token, ui/token | the MCP server's bearer token and the web console's token |
undo.json, undo/ | the undo journal and its backups |
SSH host keys are kept per environment, never in ~/.ssh/known_hosts, because re-created hosts reuse the same addresses.
Related¶
- Security model
cs help creds,cs explain ui,cs explain agentic