> ## Documentation Index
> Fetch the complete documentation index at: https://hastekit.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Codebases

> Attach GitHub repositories so an agent can read, search, and reason over code

Codebases bring source code into the agent. Connect a GitHub repository once, HasteKit keeps it synced, and any agent in the project can attach it to read files, search the tree, and answer code-aware questions.

<Frame>
  <img src="https://mintlify.s3.us-west-1.amazonaws.com/hastekit/images/agent-builder/codebases.png" alt="Codebases" />
</Frame>

## Adding a codebase

1. First, register the **GitHub connector** under **Agent Framework → [Connectors](/docs/gateway/agent-builder/connectors)**, and connect the account that has access to the repo.
2. Open **Agent Framework → Codebases**.
3. Click **Add Codebase**.
4. Select:

| Field                 | Description                                                           |
| --------------------- | --------------------------------------------------------------------- |
| **GitHub Connection** | The OAuth connection to use for cloning and syncing.                  |
| **Repository**        | `owner/name` of the repository.                                       |
| **Branch**            | Default branch (usually `main`). The codebase will track this branch. |
| **Path filter**       | Optional. Limit indexing to a subdirectory (e.g. `services/api`).     |

5. Save. HasteKit clones the repo and syncs it on a schedule (and on demand).

## Attaching to an agent

1. Open the agent → **Codebases** tab.
2. Select one or more codebases.
3. Save.

The agent receives tools for navigating and reading the repo — `list_directory`, `read_file`, `search_code`, etc. — scoped to the attached codebases. Reading is read-only; agents can't push, branch, or modify the repo.

## How sync works

* Repositories are re-cloned on a schedule (typically every few minutes) and on demand when the agent needs a fresh read.
* Only the configured branch is tracked. Switch branches by editing the codebase config.
* Large repos are paginated and chunked; the agent doesn't have to load the whole tree into context.

## Use cases

* **Code review agents** — read a PR diff, the surrounding files, and project conventions, then leave a structured review.
* **Documentation agents** — answer "where is X defined" or "what files use this function."
* **Refactor planners** — produce a step-by-step plan grounded in the actual code rather than guessed structure.

## Relation to other features

* [Knowledge Bases](/docs/gateway/agent-builder/knowledge-base) — for general docs, PDFs, markdown. Codebases are for repository-shaped data where path and structure matter.
* [Workflows](/docs/gateway/agent-builder/workflows) — GitHub trigger nodes can kick off a workflow on a PR event; the agent inside that workflow can read the same codebase.