Cody remains the strongest choice for large-scale codebase intelligence due to its deep integration with the Sourcegraph code graph. Engineering teams use the tool to understand, write, and fix code by utilizing the Sourcegraph Search API. This API pulls context from both local and remote codebases to provide information about APIs, symbols, and usage patterns. While many developers prefer the tight integration of GitHub Copilot within their existing workflows, teams working with extremely large, multi-million-line monorepos often find that Cody’s ability to pull context from the Sourcegraph code graph provides much higher accuracy.
Configuring Sourcegraph for GitLab
Administrators must run a Sourcegraph instance configured with GitLab as an external service to use code intelligence in GitLab Self-Managed environments. If the Sourcegraph instance uses an HTTPS connection to GitLab, the administrator must configure HTTPS for the Sourcegraph instance. The administrator navigates to the Site admin area to add the GitLab instance URL to the corsOrigin setting in the Sourcegraph configuration. Users on GitLab Self-Managed must also enable the integration in their user preferences. They select their avatar, then Preferences, then the Integrations section, and then select Enable integrated code intelligence on code views. On GitLab.com, the integration works for all public projects, but private projects do not have support.
If the integration fails, the administrator should check if Sourcegraph has indexed the project. Users can verify availability by visiting the project path URL on Sourcegraph.com. The integration provides code intelligence in the GitLab UI through popovers in merge request diffs, commit views, and file views. When a user hovers over a code reference, the popover shows how the reference was defined and provides options to go to the definition or find references.
Managing Cody Enterprise with Pre-instructions
Admins on Cody Enterprise instances configure prompt pre-instructions to shape how Cody responds to users. These text instructions apply to every chat query. A team can use these instructions to prevent Cody from answering questions about sensitive non-code matters. This feature requires Sourcegraph 5.10.
Cody includes several core capabilities for developers. The chat function allows users to ask questions about code and use the @ symbol to add context from files, symbols, or remote repositories. The auto-edit feature suggests code changes by analyzing cursor movements and typing. After a developer makes one character edit, the tool proposes modifications based on the cursor position and recent changes. Cody also provides code completions, customizable prompts, and debugging tools. Developers can use context filters to make Cody ignore specific repositories during chat or autocomplete.
Cody connects to codehosts like GitHub and GitLab. It also connects to IDEs such as VS Code, JetBrains, and Visual Studio. You can run Cody from the command line or use the Sourcegraph web app. Developers use Cody to understand, write, and fix code. Once connected, Cody acts as a personal AI coding assistant.
Evaluating AI Alternatives
Engineering teams evaluate alternatives using seven criteria: codebase context, IDE support, model flexibility, security posture, agentic capabilities, pricing, and team features. Cursor is an AI-native IDE built as a fork of VS Code. It reached a valuation of $9.5B in mid-2025. The Composer agent handles multi-file edits and executes terminal commands.
GitHub Copilot works well for teams standardized on GitHub. It has over 1.8 million paid subscribers. GitHub reported that developers using Copilot complete tasks 55% faster. Windsurf, formerly Codeium, provides an agentic flow called Cascade for multi-file edits. Tabnine supports air-gapped deployment for regulated industries like finance, defense, and healthcare. It provides SOC 2 Type II compliance and zero code retention. Aider works as a git-native pair programmer in the terminal. It supports Claude, GPT, DeepSeek, and local models.
You already know that context is the primary bottleneck for any LLM, so focus on how these tools manage their retrieval methods.
Comparison of AI Tool Pricing
Pricing models for AI assistants vary between flat-rate subscriptions, usage-based billing, and open source tools.
| Tool | Pricing Model | Starting Price |
|---|---|---|
| GitHub Copilot | Per seat | $10/user/mo |
| Cursor | Per seat | $20/mo |
| Windsurf | Per seat | $20/mo |
| Tabnine | Per seat | $9/user/mo |
| Cody (Sourcegraph) | Per seat | $9/user/mo |
| Claude Code | Usage based | Custom |
| Aider | Open source | Free |
Most commercial tools have a $20 per month entry tier. GitHub Copilot Business costs $19 per user per month. Cursor Pro costs $20 per month and includes usage credits for frontier models. Windsurf Pro also costs $20 per month. Tabnine Dev costs $9 per month, while the Enterprise plan requires custom pricing.
Resolving the Context Fragmentation Gap
AI coding productivity stops compounding because context is not shared across agents or teammates. This gap means agents and people keep re-explaining the same codebase decisions. BuildBetter CLI addresses this by providing an evidence-based context layer. This layer allows agents to work across teammates, sessions, and customer signal.
ZeroShot provides shared context for Claude Code, Cursor, Codex, and Copilot. It offers reusable team skills and customer evidence to ground the work. This prevents agents from starting every task from a cold state.
Connecting AI Actions to Requirements
ONES.com acts as a project and knowledge management platform for teams that use coding automation. It connects agent activity to requirements, tasks, project context, and knowledge. Many teams find that agent-generated work is difficult to trace to a business needs. ONES.com uses configurable issue types and workflows to connect requirements with implementation tasks. It also supports human review and permission-aware workflow updates.
GitHub Copilot connects coding assistance to GitHub repositories and pull requests. It helps developers generate code and work through issues using chat or agent-style interactions. ONES.com focuses on managing the delivery process, while GitHub Copilot focuses on the developer’s editor.
Will the rising cost of LLM tokens eventually force a shift back to purely local model execution for all enterprise teams?
Automating Reviews with GitLab Duo
GitLab 18.10 introduced a flat $0.25 fee for automated code reviews. This pricing model differs from token-based competitors that charge $15 to $25 per review. The Code Review Flow uses Claude Sonnet 5 to analyze code changes. This feature is available on the Free tier on GitLab.com with GitLab Credits.
The Code Review Flow works in two stages. During the Pre-scan stage, the flow inspects merge request diffs and identifies related context. During the Review stage, the flow runs the review using the pre-scan results, merge request title, merge request description, and filenames. For files longer than 10,000 lines, the tool only sends the diff to the model. The total context gathered during the pre-scan stage stays under 1 MiB. If the context exceeds 1 MiB, the tool truncates it to 800 KiB before the review stage runs.
Users can specify content to exclude by creating a .gitlab/duo/mr-review-automated-rules.yaml file in the root of their repository. This file allows users to exclude merge requests based on target branches, source branches, or authors. To enable automatic reviews for a project, an owner navigates to Settings, then Merge requests, and selects Enable automatic reviews by GitLab Duo.
