User Guide

User Guide: ContextLoom Enterprise

Version: 1.0.0 Target Audience: Software Engineers, Team Leads, AI Curators

This guide explains how to use the ContextLoom platform to create governed, high-quality context for your AI agents.

1. Core Concept: The Context Bundle

In ContextLoom, we do not simply “chat with a repo.” Instead, we create Context Bundles.

A Bundle is a curated package containing:

1. File Pointers: Specific references to files across one or more repositories (e.g., api/auth.py, frontend/src/Login.tsx).
2. System Instruction: A natural language directive telling the AI how to interpret these files (e.g., “Act as a Security Auditor reviewing this authentication logic.”).
3. Versioning Strategy: A rule defining which version of the files to serve.

Think of a Bundle as a “Mental Model” you want to transfer to the AI.

---

2. Connecting Repositories

Before you can create bundles, you must connect your VCS providers.

1. Navigate to Sidebar > Repositories.
2. Click Connect Repository.
3. Select your provider:
   • GitHub / GitLab: Enter the Owner/Repo (e.g., acme/billing-service).
   • Local: Enter the absolute path on the appliance server (e.g., /mnt/data/repos/billing).
4. Provide authentication credentials (PAT) if the repository is private.

Note: For on-premise instances (GitLab Self-Managed), ensure the Base URL is correctly configured.

---

3. The Curator Workspace

The Bundle Curator is the main interface for defining context. It is divided into two panes.

Left Pane: Source Browser

• Bundle Name: Select a name for your bundle
• Token Estimator: The bar at the bottom estimates the size of your bundle.
• Repo Selector: Choose which repository to browse.
• File Tree: Navigate the directory structure.
  • Click a folder to expand it.
  • Check the box next to a file to add it to the bundle.
  • Recursive Add: Checking a folder adds all compatible text files inside it.

Right Pane: The Manifest

This list shows exactly what the AI will see.

• Remove: Click the trash icon to remove a file from the context.
• Description: Place the description of your bundle. Avaliable for AI agents and human for searching.
• System Instruction: A natural language directive telling the AI how to interpret these files (e.g., “Act as a Security Auditor reviewing this authentication logic.”).

Pinning Strategy (Versioning)

At the top right, you will see the Version Mode toggle:

• Dynamic (Default): The bundle always fetches the latest version (HEAD) from the main branch. Use this for active development tools.
• Pinned (Locked): ContextLoom resolves the current Commit SHA for every file and locks it. Future fetches will always serve this exact version, even if the repo changes. Use this for:
  • Production regression tests.
  • Audited compliance reviews.
  • Multi-agent tasks requiring stability.

---

4. Consuming Context

Once a bundle is saved, you can use it with your AI tools.

1. Go to the Dashboard.
2. Find your bundle card (e.g., “Billing Logic v1”).
3. Click the Copy ID button. You will get a URI like cl://bundles/550e8400-e29b....

Using with Claude / Cursor

If your MCP Bridge is configured (see Client Deployment Guide):

In Prompt:

“Review the authentication logic. Use context: @ContextLoom/Billing Logic v1”

Or

“Build-up context using MCP cl://bundles/550e8400-e29b…”

You can also use prompt mcp tool /get_context

/get_context cl://bundles/550e8400-e29b…

Note: The exact syntax depends on the client integration. Some clients allow you to select “ContextLoom” from a tool menu.

---

5. User Settings & Access

Personal Access Tokens (PATs)

To connect your local tools (Claude/Cursor) to the ContextLoom Appliance, you need an API Key.

1. Go to Sidebar > Access Tokens.
2. Click Generate Token.
3. Give it a name (e.g., “MacBook Pro”).
4. Copy the token immediately. It will not be shown again.

Switching Organizations

If you belong to multiple organizations (e.g., “Frontend Team” and “Backend Team”):

1. Click organization toggler in the top left of the Sidebar
2. Choose the context you wish to work in.
   • Note: Bundles created in one organization are invisible to the other.