Getting Started

Quick start

The fastest reliable path through Calypso RAG: API key, knowledge, agent policy, Playground validation, and your first deployment surface.

Goal

This quick start gets you from an empty workspace to a real grounded answer flow using the actual Calypso RAG product path.

Before you begin

You need:

  • access to a Calypso RAG workspace
  • permission to manage the current project
  • one document or website you trust as a source of truth
You cannot send a prompt from Playground until a project API key exists and is selected.

Step 1. Create or reveal a project API key

Open Project → API Keys and make sure at least one active key exists.

This comes first because the key is required for:

  • Playground
  • Code integrations
  • the Upload API
  • n8n credentials

Step 2. Add one useful source in Knowledge

Open Knowledge and add one strong source:

  • a file with durable product or policy information
  • or a website that acts as a canonical public source

Start small. One trusted source teaches you more than a large noisy corpus.

Step 3. Wait for indexing to settle

Do not test too early. A source that is uploaded but not retrievable yet will distort your first impression of the system.

Typical states include:

  • queued
  • processing
  • active
  • ready
  • failed

Step 4. Confirm the default agent in Agent Builder

Open Agent Builder and inspect the default model:

calypso-rag-agent

Check:

  • retrieval is enabled
  • the source scope is correct
  • role and tone fit the use case
  • source visibility is intentional
  • small talk behavior is intentional

Step 5. Validate in Playground

Open Playground and confirm:

  • the selected agent is correct
  • the selected project API key is active
  • the answer is grounded in the source you just added

Good first prompts:

  • "Summarize the grounded knowledge available in this workspace."
  • "Answer as if you were helping a real customer, using only connected sources."
  • "Tell me what information is still missing for reliable answers."

Step 6. Read the result like an operator

Do not stop at "the model replied."

Check:

  • whether the answer is actually grounded
  • whether source titles are visible or hidden by policy
  • whether the retrieval scope feels too narrow or too broad
  • whether the role and presentation match the target deployment surface

Step 7. Choose one deployment path

Once Playground looks correct, move to the matching integration: