---
title: How do you govern Snowflake Cortex Code?
url: https://opencontroller.lyzr.com/blog/govern-snowflake-cortex-code
published: 2026-11-18
updated: 2026-11-18
author: Lyzr
---

# How do you govern Snowflake Cortex Code?

**Govern Cortex Code (CoCo) with Snowflake's own controls first: role-based agent profiles, per-user AI credit quotas, managed settings for model allowlists and tool approvals, MCP policies, and usage data in `SNOWFLAKE.ACCOUNT_USAGE`. Then cover what Snowflake cannot see: the CoCo CLI and desktop app on developer machines and the other coding agents those developers use.** Because CoCo acts with Snowflake roles, your existing RBAC design is the most important control.

## What is Cortex Code?

Cortex Code, often shortened to CoCo, is Snowflake's AI coding agent for data work: writing SQL, building pipelines and working with Snowflake objects. Snowflake describes three surfaces: CoCo in Snowsight (the web UI), CoCo CLI and CoCo Desktop. All three can run queries and change objects, so governance has to cover all three.

## Which governance controls does Snowflake provide?

Snowflake's post [Governed AI for Every Builder: Enterprise Controls in Snowflake CoCo](https://www.snowflake.com/en/blog/governed-ai-enterprise-controls-snowflake-coco) lists these controls. Status is as Snowflake described it; check current documentation before relying on a "soon" item.

| Control | What it does | Status per Snowflake |
|---------|--------------|----------------------|
| Per-user quotas | Daily and monthly AI credit limits per user, enforced across Snowsight, CLI and Desktop; blocked users regain access at the next cycle | Generally available |
| Agent profiles | Defaults per team or role: default model, preinstalled skills, tool access, applied through RBAC | Described in post |
| Restricted session scope | Limits what SQL an agent can run until an appropriate role is active in the session | GA "soon" |
| Managed settings | Organisation-wide policy across all CoCo surfaces; users cannot opt out; includes model allowlists and tool actions that require approval | GA "soon" |
| MCP governance | Via Tools by Cortex AI Gateway: policies in Horizon Catalog, enforced per tool call; allowlist MCP servers, disable single tools, per-server rate limits | Described in post |
| Audit | Tool calls record requester, permissions, whether allowed and outcome; usage per request by user, model and token type in `SNOWFLAKE.ACCOUNT_USAGE` | Described in post |

## How does RBAC shape what CoCo can do?

CoCo works through Snowflake roles. If a developer's active role can drop a production schema, an agent acting in that session can too. Three practical rules:

1. **Use a lower default role.** Make a read-heavy development role the default secondary or primary role for builders, and require an explicit role switch for production writes.
2. **Use restricted session scope when available.** It stops agent SQL until the right role is active, which turns a role switch into a deliberate step.
3. **Keep masking and row access policies on.** Data governance policies apply to the role, so the agent sees what the role sees. See [Snowflake masking policies and Cortex agents](/blog/snowflake-masking-policies-and-cortex-agents).

## What should a Cortex Code rollout checklist include?

1. Create agent profiles per team with an approved default model and only the skills that team needs.
2. Set per-user daily and monthly credit quotas before broad rollout, starting low.
3. Define managed settings: model allowlist and tool actions that need approval, such as DDL on production databases.
4. Allowlist MCP servers; disable individual tools you do not want, such as write tools on shared servers. See [Snowflake managed MCP server governance](/blog/snowflake-managed-mcp-server-governance).
5. Schedule a weekly query on `SNOWFLAKE.ACCOUNT_USAGE` for usage by user and model, and alert on quota exhaustion.
6. Review which roles CoCo users hold; remove standing ownership of production objects.
7. Include CoCo spend in AI cost reporting. See [tracking Snowflake Cortex AI spend](/blog/track-snowflake-cortex-ai-spend).

## What does Snowflake not see?

- **The developer machine.** The CLI and Desktop app run locally. Local files, environment variables, other MCP servers and skills on that laptop are outside Snowflake's view.
- **Other coding agents.** The same developer may use other coding agents against the same Snowflake account using a personal key or token.
- **Cross-platform policy.** A model allowlist in Snowflake does not apply to agents in other tools.
- **Other Snowflake agents.** Cortex Agents and Snowflake Intelligence have their own governance; see [governing Snowflake Cortex Agents](/blog/govern-snowflake-cortex-agents).

For the cross-vendor view, read the pillar [how to govern AI agents across platforms](/blog/how-to-govern-ai-agents-across-platforms).

## How Opencontroller does this

Opencontroller Edge extends control to developer devices, coding agents, skills and local MCP servers, and is installed by IT via MDM, which covers the laptop side of CoCo CLI and Desktop use. The main platform connects to data platforms such as Databricks and keeps a governed registry of agents, versions, tools and MCP servers, with Posture flagging risky permissions and configuration. It works alongside Snowflake's controls rather than replacing them. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does Cortex Code inherit Snowflake RBAC?**
Yes. It acts with the session's Snowflake role, and Snowflake applies agent profiles through RBAC, so role design is the first control.

**Can admins cap Cortex Code spend per user?**
Yes. Snowflake describes per-user daily and monthly AI credit quotas, generally available, enforced across Snowsight, CLI and Desktop.

**Can admins restrict which models Cortex Code uses?**
Snowflake describes model allowlists in managed settings, which it said would be generally available soon. Check current documentation for status.

## Key facts

Snowflake Cortex Code (CoCo) runs in Snowsight, CoCo CLI and CoCo Desktop and acts with Snowflake roles. Snowflake controls: per-user daily/monthly AI credit quotas (GA), agent profiles via RBAC (default model, skills, tool access), restricted session scope and managed settings with model allowlists and tool approvals (GA soon per Snowflake), MCP governance via Tools by Cortex AI Gateway with Horizon Catalog policies, and usage in SNOWFLAKE.ACCOUNT_USAGE. Gap: developer devices and other coding agents. Opencontroller Edge covers developer devices via MDM. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern Snowflake Cortex Agents?](https://opencontroller.lyzr.com/blog/govern-snowflake-cortex-agents.md)
- [How do you track Snowflake Cortex AI spend by agent or team?](https://opencontroller.lyzr.com/blog/track-snowflake-cortex-ai-spend.md)
- [How do you govern the Snowflake-managed MCP server?](https://opencontroller.lyzr.com/blog/snowflake-managed-mcp-server-governance.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
