Your first agent
Open Admin β Platform configuration β Setup Wizard. You need one reachable model and the agent runtime for your first conversation. Tools, knowledge bases, connected accounts, and automation can be added later.
When setup appearsβ
- Eligible admins see a setup checklist on a fresh installation (no conversations or custom agents, and at most one user), or while saved setup is in progress.
- The full wizard opens when the admin chooses to start or resume it.
- Completed, dismissed, or hidden setup does not prompt automatically.
- CAIPE Tour β Setup experience in the top-right bar opens the setup page for admins. Choose Start setup, Resume setup, or Run setup again there.
SETUP_WIZARD_ENABLED=falsesuppresses automatic onboarding. Manual setup and workspace tour replay remain available.
Complete basic setupβ
- Welcome: review the brief readiness summary. Open Platform Health for service inventory, component logos, detailed checks, and remediation links. A disabled optional service does not prevent a basic conversation.
- Choose a model: the model is the AI that writes your agentβs replies. The displayed choice is for this starter agent, not a platform-wide default. Keep it or open Change model to search the catalog. Registration alone does not verify provider access; the final conversation tests that connection. For a different provider, open Configure provider access and configure its endpoint and authentication in the provider workspace, then register/select the model. A model name alone does not configure access. See Provider credentials for saved-secret selection. LiteLLM uses an OpenAI-compatible endpoint. For Bedrock on Kubernetes, prefer workload identity such as an IAM role for a service account or EKS Pod Identity.
- Choose an agent: pick SRE starter, Hello World, or a blank agent. Recipes provide instructions; operational tools must be connected separately.
- Add context (optional): choose Accounts, Tools, Knowledge, or Team channels. Use Try without extras to start with a model-only conversation.
- Try your agent: review the selection and run the test. Success means the agent returned a response. If it fails, check provider access and platform health, then retry. Test later saves your place without claiming success. If a starter with that name already exists, setup creates a separately named starter from your current choices and keeps the existing agent unchanged.
Provider credentialsβ
Model Providers shows all supported providers, even before a model is registered.
- Choose Choose a saved secret for a searchable list of permitted secrets. Saving keeps the reference; it does not copy or reveal the value.
- Choose Enter a new value to create an encrypted entry in Credentials β Saved Secrets. Empty lists default to this option.
- Use Share on a saved secret to grant team access, Rotate to replace its value, or Delete and confirm to remove it. Deleting a saved key does not revoke it at the provider.
| Provider | Configuration choices |
|---|---|
| OpenAI / Compatible | Any OpenAI-compatible API base URL and API key |
| Anthropic Claude, Google Gemini, Groq | API key |
| Azure OpenAI | Endpoint, API version, and API key |
| AWS Bedrock | Deployment IAM, or access key ID, secret access key, region, and optional session token |
| Vertex AI | Deployment Application Default Credentials |
Give the connection a name and enter one or more provider model IDs. Save Connection encrypts new values in Saved Secrets and registers the models together. Existing saved values are reused unchanged. Multiple endpoints can offer the same model ID.
Use Test Connection before saving, on a saved connection, or on a model row. It runs a short completion with the same provider configuration used by agent chat. Expand the result for authentication, network, timeout, rate-limit, or model errors. A successful check does not guarantee that every tool or modality is supported.
Select the registered model in your agent. Team members need model read access and use access to each saved credential. Share credentials from Saved Secrets; rotations take effect on the next invocation and revoked access fails closed. Publishing a new endpoint/key pairing requires platform configuration access and management access to the referenced secrets. Existing deployment-managed models continue using their environment settings.
Leave and returnβ
- Use Expand setup width in the header for more room; Restore setup width returns to the compact layout without losing your step or selections.
- Save & minimize, the minimize icon, and links to CAIPE workspaces save your current step and selections before shrinking setup into the resume bubble.
- Complete the task in the destination workspace, then choose Resume setup.
- Navigation does not reopen the wizard automatically.
- Donβt show again hides the checklist. The permanent Setup Wizard page in Admin remains available, including after basic setup is complete.
- Restarting setup resets the setup checklist; it does not remove agents, credentials, models, or knowledge bases created earlier.
Enable optional featuresβ
A navigation checkbox controls visibility. It cannot install a missing service. Ask the deployment administrator to configure the feature and restart/redeploy the UI; deploy its backend where required. Then refresh setup.
| Feature | Deployment requirements | Guide |
|---|---|---|
| Workflows | WORKFLOWS_ENABLED=true, WORKFLOW_RUNNER_ENABLED=true, and a working agent runtime | Workflows |
| Schedules | Scheduler backend, authentication/runner wiring, SCHEDULER_ENABLED=true, and reachable SCHEDULER_URL | Scheduler |
| Autonomous agents | Autonomous service, its authentication configuration, ENABLE_AUTONOMOUS_AGENTS=true, and reachable AUTONOMOUS_AGENTS_URL | Autonomous agents |
| Apps | AGENTIC_APPS_INSTALL_ENABLED=true, a catalog at AGENTIC_APPS_CONFIG_PATH, and the app signing configuration | External Apps |
| Knowledge | RAG service and dependencies, ingestion, and an accessible knowledge-base MCP server | Knowledge bases |
| Team channels | A configured bot integration and its service | Slack, Webex |
Connected accounts require an enabled OAuth connector. Open Configure connected apps to set up the provider, then Credentials β Connected Apps to connect your account. Provider support determines whether dynamic client registration is available; an arbitrary MCP URL does not guarantee OAuth or DCR support.
Skip guided onboardingβ
Setup is optional. Choose Exit setup at any step to save your choices and hide the checklist across sessions without marking setup complete. Resume from Admin β Platform configuration β Setup Wizard whenever needed.
To suppress the automatic setup checklist and product tour for the deployment, set these Helm values and upgrade the release:
caipe-ui:
config:
SETUP_WIZARD_ENABLED: "false"
For the standalone UI chart, use config.SETUP_WIZARD_ENABLED: "false".
For Docker Compose, set SETUP_WIZARD_ENABLED=false and recreate the UI container.
For the installer, use --no-setup-wizard or ENABLE_SETUP_WIZARD=false.
Manual setup in Admin and tour replay in Settings remain available.
Explore the workspaceβ
Choose CAIPE Tour β Workspace tour in the top-right application header, or Take CAIPE Tour on the Setup Wizard page. The guide docks beside the workspace on wide screens and below it on smaller screens, leaving the workspace controls available. It highlights navigation while you keep using the workspace. Use Next and Back to explore Home, Chat, Skills, and personal settings. The tour opens each stepβs page automatically, including the correct workspace tab. Existing chats and forms on that page stay open. Unsaved drafts use the usual discard confirmation; canceling leaves the draft in place.
Minimize the guide to a Resume tour button, or close it whenever you like. Progress is saved to your account. Replay it from the header or Settings. Deployment details and optional navigation capabilities stay collapsed until you request them; setup keeps the next decision in focus.
Task toursβ
The top-right CAIPE Tour menu includes short guides for:
- Add a model: connect its provider, register identifiers, and verify with chat.
- Add an MCP server: configure the endpoint and authentication, then test a tool.
- Add an agent: choose a model and audience, write instructions, attach tools, and try it.
- Chat with an agent: select an agent, ask a question, inspect evidence, and follow up.
- Connect credentials: authorize a connected app, check its health, or save a secret.
Each task tour saves its own progress to your account and starts only when requested. Choosing an unfinished task resumes it; choosing a completed or skipped task starts it again. The agent guide follows the builder automatically: use the formβs Next, Previous, and Create Agent buttons. Other tours use the guideβs Next after each action. Finish tour records guide completion and offers Back to Home or Stay here. Returning Home also respects unsaved drafts. Finishing a tour does not verify or create the resource. Model, MCP, and agent guides appear when the agent workspace is enabled and your account has access. Connected credentials guidance appears when account connections and persistent storage are enabled. The underlying screens enforce the permissions required for each action.
CAIPE Tour controlsβ
- Settings β Chat & agents β CAIPE Tour: turn Enable CAIPE Tour off to hide the menu and stop guided tours for your account across browsers. Turn it back on there; saved progress is retained.
- Admin β Platform configuration β Tour β CAIPE Tour: use Enable CAIPE Tour to turn tours on or off for everyone at runtime. The switch saves immediately and retains selected tours and saved progress. Check the tours users may see, then choose Save tours. Permissions and deployed features still limit availability.
- Existing deployments show all tours until an operator saves a selection. Preferences and the operator selection are stored in MongoDB; no redeploy is needed.
- Policy changes apply immediately to API requests. Other browser sessions update on focus or within a minute. An operator change cannot re-enable a userβs opt-out.