Sourcegraph Cody

What is Sourcegraph Cody?

Sourcegraph Cody is an AI-powered programming assistant that understands code, developed by Sourcegraph. It combines large language models with Sourcegraph’s code search, symbol navigation, and Code Graph functions to retrieve context from local files, remote repositories, and cross-repository dependencies, thereby assisting developers in understanding, writing, modifying, and explaining code.

In its early days, Cody offered Free, Pro, and Enterprise versions for both individuals and businesses; however, its product orientation has now changed significantly: the individual and small-team versions are no longer available, and Cody is maintained only as an enterprise-level version of Sourcegraph. When looking for information on Cody, it is necessary to distinguish between historical tutorials and the current enterprise product.

Current status: The personal version is no longer in service.

On June 25, 2025, Sourcegraph officially announced that it would discontinue the Cody feature in versions Cody Free, Cody Pro, and Enterprise Starter. New users could no longer register for the Free or Pro versions starting from the date of that announcement, and users affected by this change lost access on July 23, 2025; the Enterprise Starter version would also retain only Code Search, without the Cody feature anymore.

Cody Enterprise has not stopped operating.Sourcegraph states that it will continue to provide support and investment for the enterprise version.

For individual developers who previously used Cody Free or Pro, the new tool recommended by the developers is Amp, an AI coding tool that emphasizes agent-based and collaborative development. The concepts mentioned in older articles such as \"Cody Free version\", \"Cody Pro monthly subscription\", and the number of free model uses are no longer valid.

Core functions

1. Code repository chat

Cody can answer questions regarding code structure, module responsibilities, call relationships, causes of errors, and implementation methods. It selects relevant files through semantic search, Sourcegraph Search, and Code Graph; users can also specify the context using files, directories, or symbols.

Compared to ordinary chat systems that only read the currently open files, Cody’s advantage lies in its ability to make use of the remote repositories that have been indexed by Sourcegraph; it is particularly suitable for large monolithic repositories, microservices, and multiple code hosting platforms. However, dynamic configurations, runtime dependencies, or unindexed resources may still go unnoticed during searches, and the answers provided must be verified by examining the actual code and tests.

2. Cross-warehouse context

Cody Enterprise enables access to context from multiple repositories, allowing developers to ask how shared libraries are invoked by different services, which projects will be affected by changes in a particular interface, or to understand the API of a repository even when it has not been cloned locally. This multi-repository functionality is built on Sourcegraph’s search indexing and permission synchronization mechanisms.

The more warehouses there are, the more important permissions and data boundaries become. Cody should only be able to see the code to which the user is authorized to have access;

The administrator also needs to check the synchronization of code host permissions, the index range, and the context sent to the model.

3. Automatic code completion

Cody provides single-line and multi-line completion as well as automatic editing suggestions in supported clients such as VS Code, JetBrains, and Visual Studio. The model uses the current file, the content surrounding the cursor, and the context of the codebase to suggest functions, tests, configurations, or recurring patterns.

Patches are applied directly to the source code, which can lead to the inclusion of non-existent APIs, insecure practices, and licensing issues. Developers should review each patch, running formatting, type checking, testing, and security scans.

4. Inline Edit and Smart Apply

Users can select code and request that it be fixed, refactored, explained, tested, or for documentation to be added. VS Code’s Smart Apply feature allows chat suggestions to be applied to files;

Different editors have varying levels of functionality; JetBrains, Visual Studio, Web, and CLI may not support the same methods of use.

Major modifications should be carried out in separate branches. Automatic application tools cannot take into account all business constraints, especially those related to database migration, concurrency, authentication, and changes in cross-service protocols.

5. Prompt Library

Cody offers reusable prompts for explaining code, generating unit tests, writing documentation, and handling common team tasks. Companies can promote standardized prompts to enable developers to carry out reviews or migrations in a consistent format.

The prompt is not a policy executor; the output still needs to be verified. Keys, internal tokens, or unnecessary sensitive information should not be included in the prompts.

6. Multi-model configuration

Cody Enterprise allows administrators to choose between chat, quick chat, and code completion models.

Models can be provided through the Sourcegraph Model Provider and Cody Gateway; it is also possible to use BYOK to connect to Anthropic, OpenAI, Google, Amazon Bedrock, GCP Vertex, or internal endpoints that are compatible with OpenAI, and self-hosted models can be configured as well.

Administrators can set which models are allowed or prohibited, as well as the publication status, default models, and context windows. Models obtained through Gateway are subject to the context constraints defined by Sourcegraph.

When more context is needed, BYOK can be used along with the appropriate window configuration.

The models and versions are continuously updated; the model configuration provided by the enterprise instance should be used as the reference.

7. Context Filters

Enterprise administrators can use warehouse name rules to determine which warehouses can be used as Cody contexts, and they can prevent sensitive data from those warehouses from being sent to third-party models. Include, exclude, or combined rules can be configured, and these rules are applicable to supported VS Code, JetBrains, Visual Studio, Web, and CLI clients.

By default, when no filtering is configured, Cody can access all repositories within its permission scope as context for third-party model requests. It is necessary to define these boundaries explicitly before deployment, rather than assuming that sensitive repositories will be excluded by default.

The client and the Sourcegraph instance must also meet the minimum version requirements.

8. Management, RBAC, and Analysis

Cody Enterprise allows control over which users can use AI through Sourcegraph roles; administrators can also set the validity period of IDE access tokens and view usage analytics. Enterprises can integrate access controls with code host permissions, SSO, and organizational security procedures.

Analyzing data is useful for understanding the adoption rate and usage levels; it should not be sufficient to evaluate an engineer’s performance based solely on the number of completed tasks or the amount of communication. High usage levels do not necessarily indicate high quality, and low usage can also be due to differences in the type of tasks.

Supported editors and clients

The main clients used for comparing documents currently include VS Code, JetBrains, Visual Studio, Sourcegraph Web, and CLI. There are differences in terms of capabilities such as chat, context awareness, completion, and automatic editing:

• VS Code：It offers the most comprehensive set of functions, including chat history, context for files/symbols/directories, Smart Apply, image input, model selection, completion, and automatic editing.
• JetBrains：It supports chatting, multi-warehouse context, completion, and automatic editing, but some of the Smart Apply functions and symbol features are different;
• Visual Studio：It supports chatting, symbol context and completion, but lacks some advanced features.
• Web：Suitable for chatting and searching in remote repositories; no local auto-completion is provided.
• CLI：It offers chat and local Agent-based context, but lacks a complete chat history as well as an interface for selecting multiple repositories.

Neovim extensions are in an experimental stage, are no longer actively maintained, offer limited support, and therefore are not suitable as the primary interface for new enterprise deployments. Cody can also be accessed via instance URLs and access tokens in VS Code-derived editors such as Cursor, Gitpod, and IDX.

Code host and enterprise deployment

Sourcegraph can be connected to major code hosting services such as GitHub, GitLab, Bitbucket, Azure DevOps, Gerrit, and Perforce, and it supports the use of multiple code sources. Cody can be enabled either through Sourcegraph Cloud or via the self-hosted Version Control solution Sourcegraph Enterprise.

Self-hosting does not automatically mean that the model is also located locally. If Cody Gateway or an external model provider is used, the filtered code context will still be sent to the corresponding service.

To achieve stricter data residency, BYOK, internally compatible endpoints, or a self-hosted model are required, along with the verification of the entire network path.

Price and purchase methods

Cody no longer has separate pricing options for individual users in the Free or Pro tiers. For Cody Enterprise, it is necessary to contact Sourcegraph’s sales team; it is configured as an enterprise-level platform, and the official party does not provide a fixed, independent pricing for each user license of Cody.

The current starting price for Sourcegraph’s Enterprise platform is16,000 dollarsIt includes AI-related features such as Credits, and these capabilities expand as the size of the team grows. The platform also offers functions like Code Search, Deep Search, Batch Changes, Insights, Monitoring, MCP, as well as API and CLI tools; the amount of 16,000 dollars should not be construed simply as the price of Cody alone.

Enterprise Credits are shared among users within an organization; the official website states that they remain valid on a monthly basis, can be carried over upon renewal, and it is possible to purchase capacity packages. The actual details regarding contracts, seats, deployments, models, support services, and the amount of Credits required are determined through a sales quote.

Fees are also charged by the model manufacturer when a model key is provided.

What is the difference between Cody and Amp?

Cody Enterprise is focused on understanding codebases, providing code completion, enabling chat functionality, managing permissions, selecting models, and taking advantage of the Sourcegraph context in large organizations; Amp is designed for modern agent-based development, placing greater emphasis on autonomous execution, collaboration, and end-to-end tasks – it represents the official path for individual Cody users to make the transition.

It’s not a simple matter of just changing the name. When a company already has Sourcegraph indexing, the ability to search across different repositories, and meets certain security requirements, Cody remains useful.

When individual developers are looking for coding agents that can be registered directly, they should consider Amp or other existing personal products.

Security, privacy, and code handling

Sourcegraph has explained to Cody Enterprise that no data is retained, and that models are not trained using customer data; it also offers compliance with SOC 2 Type II, GDPR, and CCPA standards. Enterprises can further control data through Context Filters, RBAC, token expiration periods, BYOK, and self-hosting options.

The model provider, the method of deployment, and the version can affect the actual data flow. When making a purchase, it is necessary to check the DPA, sub-processors, logs, retention period, training policies, location, and deletion mechanisms.

Using BYOK does not automatically resolve all issues; keys, model endpoints, and gateway logs also require proper management.

The public code Guardrails mentioned in the official documentation has been marked as Deprecated and is no longer recommended for use. It works only with specific combinations of Gateway and client, and it does not support BYOK.

Teams should not rely on this feature to automatically handle open-source licenses and code ownership; instead, they should continue to use SCA, license scanning, and manual review.

Open-source status and GitHub

Sourcegraph has made the Cody code repository available on GitHub, under the Apache-2.0 license. This repository includes components such as VS Code, JetBrains, Agent, CLI, and Guardrails, allowing developers to build and explore client applications.

The open-source nature of the client side does not mean that all enterprise-level services are available for free. Sourcegraph’s cloud indexing, Cody Gateway, commercial licenses, enterprise management features, model quotas, and support are still provided in accordance with the terms of the product.

To run the client, an available Sourcegraph instance and model services are also required.

The warehouse’s README may still contain outdated information regarding Free and Pro versions; it is necessary to rely on the official announcement about service discontinuation and the current documentation.

Sourcegraph Cody usage tutorial

Complete a basic task.

1. Define the code tasks to be carried out in Sourcegraph Cody, the scope of the repositories, and the acceptance criteria.
2. Connect to or import the test project, and first back up the current branch;
3. First, use the code library chat to generate a plan, and then confirm the files that need to be modified;
4. Make minor changes using the cross-warehouse context;
5. Run tests, static checks, and builds; unverified code is not accepted directly.
6. Manually check permissions, keys, dependencies, and handle exceptions before merging;

Create reusable professional workflows

1. Select a low-risk, real-world project as a template;
2. Fix the order of using code library chat, cross-repo context, and code autocompletion;
3. Record the environment, model, prompts, and failure conditions;
4. Set up manual approval for write, deploy, and delete actions;
5. Compare speed, cost, test pass rate, and the amount of rework;
6. Expand to a team or production environment only after stability has been verified;

Which users is it suitable for?

• Large development organizations that possess a large number of warehouses, microservices, or multi-code hosts;
• Engineers who need to understand code across different warehouses, conduct searches, and use AI for answering questions;
• Companies that wish for the administrator to unify the models, permissions, context, and quotas;
• Teams that need to provide a unified AI assistant in VS Code, JetBrains, and Visual Studio;
• Organizations with sensitive security requirements that use Sourcegraph Cloud, self-hosted solutions, or custom model endpoints;
• Customers who have purchased Sourcegraph Enterprise and wish to utilize code graph context for AI.

Product advantages

• It integrates Sourcegraph Search and Code Graph deeply, rather than relying solely on the current file;
• It supports remote and cross-warehouse contexts, making it suitable for complex enterprise codebases;
• Administrators can choose from the Gateway, BYOK, and self-hosted models as needed;
• Context Filters, RBAC, and token policies provide enterprise-level control;
• It supports a variety of code hosts, deployment patterns, and popular IDEs;
• The Cody client repository is open-source under Apache-2.0.

Limitations and precautions

• Individual users can no longer register for CodyFree or Pro, and installing older extensions does not guarantee access to available services.
• The free usage limit mentioned in third-party articles, the monthly fee of $9, and the list of personal models are no longer valid.
• The quality of Cody’s responses depends on the index, permissions, search results, and the model selected;
• In large codebases, missed errors, misleading legacy code, illusions, and incorrect modifications can still occur; the generated code must be reviewed and tested.
• The initial cost for setting up such a company is high; it is suitable for organizations that already need the Sourcegraph platform, but not for small teams that only want to acquire a single AI completion service.
• When making a purchase, it is necessary to take into account the costs associated with the Sourcegraph platform, Credits, model BYOK fees, as well as deployment, operation, and support costs.
• Some documents cover multiple versions, and the documentation for Guardrails, Neovim, and older models may be obsolete.
• Before upgrading Sourcegraph and the IDE plugins, administrators should check the version compatibility matrix and verify it on a test instance.

Frequently Asked Questions

Can Cody still use it for free?

No. Cody Free was discontinued on July 23, 2025, and Cody Pro was also taken off line at the same time.

Now only Cody Enterprise continues to provide support.

How much is Cody Pro?

Cody Pro has been discontinued, so it’s no longer appropriate to refer to the historical monthly fees. For the enterprise version, contact sales; the current public pricing for the entire Sourcegraph platform is 16,000 dollars.

What should individual users use?

Sourcegraph officially designates Amp as the path for Free and Pro users to make the transition. Amp is not the same product as Cody; its pricing and working principles need to be examined separately.

Which IDEs does Cody support?

It primarily supports VS Code, JetBrains, Visual Studio, the Web, and CLI. Neovim extensions are experimental and are no longer actively maintained.

Can Cody understand multiple warehouses?

Yes. Cody Enterprise can retrieve information from remote, multi-repository environments using Sourcegraph indexing and code graphs, provided that the instances are connected to those repositories and authorized to access them.

Can I use my own model?

Yes. Administrators can configure Anthropic, OpenAI, Google, cloud-hosted services, or internal endpoints that are compatible with OpenAI; they can also set up some self-hosted models.

Is Cody open source?

The Cody client repository is available under the Apache-2.0 open-source license, but the full Sourcegraph enterprise services, Gateway, model quotas, and management platform are not available for free as part of this open-source framework.

Has Cody Enterprise stopped operating?

No. The services that are discontinued are Cody in the Free, Pro, and Enterprise Starter plans.

Enterprise customers continue to receive support and development efforts.