Enablement intro

What streamlit-coco is#

streamlit-coco is a Python/Streamlit library that embeds Snowflake's Cortex Code Agent SDK ("CoCo") into a Streamlit app: a right-rail or full-panel Copilot, human-in-the-loop approval gates for destructive tools, a live preview of an app CoCo just wrote, structured output into your own widgets, and a headless query() / CocoSession API for scripts and CI.

PyPI package: streamlit-coco. Current version: 0.1.9 (alpha — the API may still move). Python 3.10+, Streamlit 1.53+. License: Apache-2.0.

The problem it addresses#

Wiring the CoCo SDK into a Streamlit app by hand means building, from scratch: session state that survives Streamlit reruns, streaming render of the agent transcript, readable tool cards instead of raw JSON tool dumps, and an approval flow for anything the agent could write, edit, or execute. streamlit-coco provides that plumbing as a set of composable functions (panel(), chat_input_bar(), copilot_rail(), render_approvals(), …) so an app author writes the page and the options, not the session bridge.

Who this is for#

Consultants and engineers who need to add a CoCo-based Copilot to a Streamlit app on a Snowflake account with Cortex Code access. It assumes basic Python and enough Streamlit familiarity to read st.container / st.session_state code — the library hides most of the session-management plumbing, so deep Streamlit expertise is not required.

Learning objectives#

After this track you will be able to:

1. Explain what streamlit-coco does, and when it applies versus using the

   CoCo CLI or the SDK directly (see the README's "When not to use" section).

2. Install the library, connect it to a Snowflake account, and run each

   bundled example app (make chat, make backlog, make bi-semantic, `make headless`, …).

3. Build a Streamlit app using the preferred pattern — panel() +

   chat_input_bar() — with require_approval_for configured so destructive tools pause for a human decision.

4. Configure CocoOptions for a real workflow and embed copilot_rail()

   into an existing multipage app.

5. Diagnose a session that fails to connect or a tool that will not run,

   using check_environment(), STREAMLIT_COCO_DEBUG, and the Technical Guide's help reference.

Time estimate#

The training cases (01-simple, 02-intermediate, 03-advanced) are a separate track, budgeted 20–30, 45–60, and 30–45 minutes respectively.

Prerequisites#

• Python 3.10+ and uv installed locally.
• A Snowflake account with Cortex Code access, and a working

  ~/.snowflake/connections.toml (legacy config.toml also works) — or the Snowflake CLI's default connection.

• The CoCo CLI (cortex) installed and reachable on PATH (verify with

  cortex --version), or a path you can set via CORTEX_CODE_CLI_PATH.

• Git access to the development repository

  (github.com/DevoteamSP/streamlit-coco-dev) if you plan to run the bundled examples, or just pip install "streamlit-coco[sdk]" if you only need the published package.

• Nothing beyond that: no dedicated Snowflake warehouse is required to work

  through this track — the examples default to read-only sample data or disconnected fixtures.