Software & Apps

Cody agents move into the enterprise

Sourcegraph provides three specialized Cody agents designed to handle business workflows like proactive visitor routing, internal knowledge retrieval, and customer support. These agents offer flexible deployment across cloud and on-premises environments while maintaining model-agnostic control.

Cody agents move into the enterprise

Specialized agents for business workflows

Sourcegraph provides three specialized Cody agents to handle specific business needs. The proactive agent analyzes visit history and page views to open conversations that match visitor intent. This agent recognizes intent, starts conversations at the right moment, and routes qualified visitors toward the next step or a meeting. It converts reads by looking at visit history and the pages a user views. The proactive agent captures context to route the right buyer straight into a meeting. The knowledge agent turns internal policies and playbooks into clear answers with the source attached, which prevents adding extra tickets to the queue. For customer-facing needs, the support agent provides grounded answers on websites so teams focus on conversations that need a human. These agents call systems, resolve live values, and complete multi-step work with user approvals. Each agent works with knowledge, tools, permissions, and deployment control. The decision to add a second agent is a configuration choice because all agents inherit the same knowledge and integrations.

Deployment and security control

Deployment for Cody agents remains flexible in cloud, private cloud, hybrid, or on premises environments. Most teams pick one problem and put an agent in front of real users. However, the company requires a Solution Design Study for complex or regulated deployments, which means the team must map the data, the controls, the integration surface, and the rollout before any implementation work begins. This mandatory mapping process prevents the immediate deployment that simpler tools allow. Cody stays model-agnostic, so teams choose a model for each workflow and switch without rebuilding the pipeline. Role-specific scope follows each user, so every person retrieves only what they have the permission to see. The enterprise implementation allows for deployment in cloud, private cloud, hybrid, or on premises environments to ensure data stays inside the environment.

Feature Capability
Chat Ask questions, generate, and edit code
Auto-edit Propose modifications based on cursor position
Prompts Automate tasks with customizable instructions
Context Pull data from local and remote codebases
Debug Identify and fix errors in code

Technical context and integration

Cody works in VS Code, JetBrains, Visual Studio, and the web app. The extension installation is straightforward because you likely already use one of these environments. Cody works with codehosts like GitHub and GitLab. Cody uses the advanced Sourcegraph Search API to pull context from both local and remote codebases so that you can use context about APIs and symbols from across your entire codebase. This integration allows Cody to use context from documents and databases. Users can also use context filters to ignore specific repositories from chat and autocomplete results. Chat capabilities allow you to ask questions about your code using @ to add context on specific files, symbols, remote repositories, or other non-code artifacts. Chat also includes the context of your open file and repository by default. You can use Cody’s chat to ask questions about your codebase. When you run a search query and open a repository, you will find the Cody button that takes you to the chat interface. The developer chat uses the most powerful models and context to help you understand, write, and fix code faster. Auto-edit functions suggest code changes by analyzing cursor movements and typing. Once you make at least one character edit in your codebase, the system begins proposing contextual modifications based on your cursor position and recent changes. You can also run Cody from the command line.

Does the model-agnostic design mean developers must learn new prompting styles for every switch? For individuals using Cody via Sourcegraph.com, Sourcegraph collects prompts and responses to improve the user experience.