Context

Calypso Context

The knowledge layer behind Calypso — documents, data, websites and images in buckets, answered from with a citation on every reply.

Calypso Context is everything Calypso knows about your business. You load sources into buckets, point an agent at the buckets it is allowed to read, and ask. The answer comes back with the source that produced it, and nothing outside those buckets is used.

Today Context is reached mainly through the API. If you run WhatsApp conversations on Smart Boards, the same buckets appear there as a board's knowledge — see Calypso WhatsApp for the dashboard side, and Knowledge on a board for how the two connect.

The four moving parts

PartWhat it is
SourceOne thing you loaded: a PDF, a CSV, a web page, a product photo.
BucketA named collection of sources (support-handbook, pricing) that agents search as a unit.
AgentA saved policy: which buckets it may read, how it answers, whether callers see sources.
AnswerGrounded output plus the citations that back it.

Every ingestion route is bucket-addressed — the destination is in the URL, so it is impossible to load knowledge "nowhere":

POST /v1/buckets/{bucket}/files

What it reads

Context indexes four families of source. Exact MIME types, extensions and per-type behaviour are in Files.

Documents

PDF, Word, Markdown, and plain text. PDFs are read visually as well as textually, so a citation can point at the page it came from.

Data files

CSV and JSON, plus source code. Loaded as text — export a spreadsheet to CSV first.

Websites

Point at a URL and Calypso reads the page. See Websites.

Images

PNG and JPEG, read for what is in the picture rather than the file name, and cited as media.

How it is kept

  • Scoped per agent. An agent answers from the buckets it is bound to. A support agent reading the returns policy does not see the pricing bucket. See Agents.
  • Traceable in every answer. Each reply carries the source it came from. A per-agent policy decides whether the caller sees those sources or only your team does. See Citations and sources.
  • Scoped per key. A project API key resolves one workspace's buckets and agents, and can be restricted further. See Capabilities.

Where to start

Quick start

Get a key, load a source, and make your first grounded call.

Ingest

Every route that puts knowledge into a bucket, and the one readiness bit to wait on.

Ask

Responses and Chat Completions, through the OpenAI SDKs you already use.

Deploy

The same agent through MCP, the Upload API, n8n, or the web widget.