Skip to content

AI‐enabled Codespace

Tyler Butler edited this page Mar 19, 2026 · 2 revisions

AI-enabled Codespace

The AI-enabled devcontainer profile provides a Codespace pre-configured with AI CLI tooling on top of the standard Fluid Framework development environment.

Creating the Codespace

  1. Go to the repository on GitHub and open the Create codespace flow. Don't just click the + button - that will use the Standard codespace image. Instead, create a custom codespace. Eventually this image will replace the Standard image.
  2. Choose your branch.
  3. Expand Dev container configuration and select AI-enabled. The profile defaults to 32 CPUs / 64 GB RAM.
  4. Create the Codespace.

First-time authentication

After creating a new AI-enabled Codespace, you may be prompted to authenticate up to a dozen times. A dozen is not a typo. This is excessive but expected — just keep clicking through each prompt until they stop.

Differences from the Standard profile

The AI-enabled profile includes everything in the Standard profile, plus:

Addition Purpose
Agency CLI Run AI agents (Claude, GitHub Copilot) from the terminal.
Repoverlay Overlay system that adds context files (agents, skills) to the workspace.
GitHub CLI (gh) Pre-installed for PR workflows and gh codespace ssh.
SSH daemon Enables gh codespace ssh access so you can connect from a local terminal.
ADO Codespaces Auth extension Authenticate to Azure DevOps from within the Codespace.
Agent aliases Shell aliases for common AI commands (see below).
Higher compute defaults 32 CPUs / 64 GB RAM vs. 16 CPUs / 64 GB RAM for Standard.

Installing Agency manually

The automated agency install during container build does not work yet. Run this command manually after your Codespace starts:

pnpm install:agency

Agency is added to your PATH automatically in all shell sessions once installed.

Agent aliases

The AI-enabled profile installs shell aliases that are available in every terminal session. All aliases route through agency.

If you're not sure what to use, try copilot-ado for feature work.

Claude aliases

Alias Command Purpose
claude agency claude Start Claude Code with the default model.
haiku agency claude --model haiku Start Claude Code with the Haiku model (fastest, cheapest).
sonnet agency claude --model sonnet Start Claude Code with the Sonnet model (balanced).
opus agency claude --model opus Start Claude Code with the Opus model (most capable).

Copilot aliases

Alias Command Purpose
copilot agency copilot Start GitHub Copilot in the terminal.
copilot-ado agency copilot --mcp 'ado --org fluidframework' Copilot with the Azure DevOps MCP server connected to the fluidframework org.
copilot-kusto agency copilot --mcp 'kusto --service-uri https://kusto.aria.microsoft.com' Copilot with the Kusto MCP server for telemetry queries.
copilot-oce repoverlay switch ff-oce && copilot-kusto Switch to the ff-oce overlay, then start Copilot with Kusto. Designed for OCE (On-Call Engineer) workflows.
copilot-work agency copilot --mcp 'workiq' Copilot with the WorkIQ MCP server.

Connecting via SSH

You can connect to a running AI-enabled Codespace from your local terminal using the GitHub CLI:

gh codespace ssh

This is useful for running AI agents from your local terminal while the Codespace provides the compute and repository context.

Clone this wiki locally