# AI Coding Agent > Part of the NocoDB documentation (Agents). Index of all pages: https://nocodb.com/llms.txt. Any docs page is available as Markdown by adding `.md` to its URL. URL: https://nocodb.com/docs/agents/coding-agent Last updated: 2026-10-29 The NocoDB AI coding agent is a cloud coding agent for GitHub: it works in the background on its own branch and machine, and opens pull requests to review. Create a coding agent that fixes [reported bugs] and opens a pull request Create a coding agent that keeps [the dependencies] of our repository up to date The NocoDB coding agent is a cloud coding agent: an AI coding agent that works in your GitHub repositories, in the background. You give it a task: an issue, a pull request, or a description. The agent starts a session on its own machine. It makes the change on its own branch and runs the tests. When you ask it to, or when its instructions say so, it pushes the branch and opens a pull request. While the agent works, you can follow the chat and see a live screen and a preview of the app. You can also review the changes and use a terminal. A coding agent is a NocoDB [agent](/docs/agents), so it also has instructions, access, triggers, channels and skills. A coding agent also has two more settings sections: **Code** and **Machines**. ## Work with the coding agent ## How does the coding agent work? The NocoDB coding agent works in runs. A coding run has these steps: 1. You start a session with a task, or a [trigger](/docs/agents/automate-coding) starts one. 2. NocoDB starts a fresh copy of the agent's [machine](/docs/agents/machines) and clones the repository. 3. The agent reads the project. It also reads the instruction files at the root of the repository: `CLAUDE.md`, `AGENTS.md` or `.claude/CLAUDE.md`. 4. The agent works on its own branch. NocoDB names the branch with the [branch name](/docs/agents/repositories#branch-name) pattern of the agent. 5. The agent makes the change and runs the tests and the checks of the project. 6. When you ask it to, or when its instructions say so, the agent commits and pushes the branch. It then opens a pull request. See [Pull requests](/docs/agents/repositories#pull-requests). 7. You review the changes. To ask for changes, comment on lines in the session or send a message. See [Coding sessions](/docs/agents/coding-sessions). The agent does not commit, push or open a pull request on its own initiative. A coding agent that you [set up](#set-up-a-coding-agent) from the **Agents** page has instructions that tell it to open a pull request. Each run starts on a fresh copy of the machine, so runs do not affect each other. ## Run coding agents in parallel The NocoDB coding agent works like a small software factory: each task gets its own session, branch and machine. Sessions do not affect each other, so the agent can work on several tasks at once. * One coding agent can have up to 3 triggered coding runs at the same time. * A project can have more than one coding agent, for example one for bugs and one for dependency updates. * [Triggers](/docs/agents/automate-coding) start runs on a schedule or when records change, without a chat. Each run ends in a branch and a pull request that a person reviews. ## Before you begin * To create a coding agent, you need the **Creator** role or higher in the project. * Coding agents are available on paid NocoDB Cloud plans. See [Availability](#availability). * Coding agents work with GitHub repositories. You need permission to install a GitHub App on the repositories that the agent works in. In a GitHub organization, an organization owner may have to approve the installation. ## Set up a coding agent You set up a NocoDB coding agent from the **Agents** page, in three steps. 1. Open the project. 2. In the mini sidebar, click **Agents**. 3. Click **Coding agent**. The **Set up your coding agent** steps open: **Repositories**, **First task** and **Launch**. ### Step 1: Pick the repositories The **Where should it work?** step connects GitHub and lets you pick the repositories. The agent sees only the repositories that you pick. 1. If the project has no GitHub connection yet, click **Connect GitHub**. A GitHub window opens. NocoDB shows **Waiting for GitHub…**. 2. In GitHub, install the NocoDB GitHub App. 3. Give the app access to the repositories that the agent works in. 4. Go back to NocoDB. The step updates on its own and shows **Connected to GitHub as** and your GitHub account or organization. 5. Select one or more repositories. You can pick up to 10 here, and add more later in the settings. 6. Click **Continue**. Step 1 of Set up your coding agent, connected to GitHub, with a repository selected The first repository that you pick is the **Default** repository. The first task runs there. When the project has more than one GitHub connection, click **Switch** to pick another connection. When no repositories show, the GitHub App has access to none yet. In GitHub, give the app access to some repositories. Then come back to NocoDB. This step shows only the GitHub connections that the project can use. A connection that you make here is restricted to this project. A connection that is restricted to other projects does not show. To use it, give this project access to the connection. See [Manage integration access](/docs/product/integrations/actions-on-connection#manage-integration-access). ### Step 2: Give it a first task The **Give it a first task** step asks for something small. You watch the agent work. Then you review the pull request. 1. Select a task from one of the tabs: | Tab | Shows | | ----------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Suggested** | Up to 3 tasks. Good first issues come first, then pull requests with changes requested, then bugs, then the most recently updated. NocoDB does not suggest draft pull requests. | | **Issues** | The open issues of the default repository, recently updated first. | | **Pull requests** | The open pull requests, recently updated first. | NocoDB selects the first task of the tab. On **Suggested**, click **Browse all** to open the **Issues** tab. 2. To give a task of your own, type it in the search box. 3. Select **New branch and draft PR**. 4. Click **Continue**. Step 2 of Set up your coding agent with the Suggested, Issues and Pull requests tabs ### Step 3: Launch The **Launch** step names the agent and starts its first run. 1. Type a **Name** for the agent. The default name is "Issue fixer". 2. Click **Start first run**. NocoDB then does these steps: 1. It creates the coding agent and adds the repositories. 2. It looks for the tools that the default repository needs. When it finds some, it creates a machine with them and builds it. When the build cannot start, the first run uses the base machine. 3. It starts the first session on the task. 4. It asks the agent to open a draft pull request when the work is done. 5. It opens the session, so you can watch the agent work. The instructions, machine, tools and skills start from defaults. You can change all of them later in the agent settings. ## Default instructions A coding agent that you set up from the **Agents** page starts with instructions for issues. The agent reproduces the issue with a test that fails and fixes it with the smallest change. It then runs the tests and lint, and opens a pull request that links the issue. To change these instructions, see [Instructions and files](/docs/agents/instructions). ## Instructions from the repository The coding agent reads `CLAUDE.md`, `AGENTS.md` or `.claude/CLAUDE.md` at the root of the repository. It follows these files for how to build, test and style the code. Keep the commands and conventions of your project there, so every session starts with them. The agent treats these files as content from the repository, not as instructions from you. It does not follow them to send code, tokens or environment values anywhere. ## Coding agent settings A coding agent has all the [agent settings](/docs/agents/instructions), plus two sections: | Section | What you set | | ------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Code** | The repositories, the branch that runs start from, the branch name, and how the agent opens pull requests. See [Repositories and code settings](/docs/agents/repositories). | | **Machines** | The machines that sessions run on: languages, databases, tools, secrets and size. See [Machines](/docs/agents/machines). | In **Tools & access**, the **GitHub** row is always on. The agent can also use the other rows, for example to read records about a bug or to look up documentation on the web. See [Tools and access](/docs/agents/tools-and-access). Some agent settings work differently for a coding agent: * The **Triggers** section has no **Results go to** setting. The result of a coding run is the session and the work on its branch. * A [channel](/docs/agents/channels) message does not start a coding session. Start sessions from the agent page or with a [trigger](/docs/agents/automate-coding). ## Frequently asked questions ### Does the coding agent change the default branch? No. The NocoDB coding agent works on its own branch. When a run starts from the default branch, the agent creates a new branch from it. The agent opens a pull request for your review. See [Repositories and pull requests](/docs/agents/repositories). ### Which repositories can the coding agent see? The NocoDB coding agent sees only the repositories that you pick, and only when the NocoDB GitHub App has access to them. See [Step 1: Pick the repositories](#step-1-pick-the-repositories). ### Does the coding agent work with GitLab or Bitbucket? No. The NocoDB coding agent works with GitHub repositories. ### What does a coding run cost? A coding run uses the [credits](/docs/product/workspaces/credits) of the workspace for its AI model, by token, and for the time its machine runs. A run on your own API key does not use credits for the model. See [Availability](#availability). ### Can the coding agent run on a schedule? Yes. Add an **At scheduled time** trigger to the coding agent. See [Automate the coding agent](/docs/agents/automate-coding). ## Availability NocoDB coding agents are available on these plans: | | Coding agents | | ------------------------------------------------------------ | ------------- | | **NocoDB Cloud, Plus, Business, Scale and Enterprise plans** | Available | | **NocoDB Cloud, Free plan** | Not available | | **Self-hosted NocoDB** | Not available | When the plan of the workspace no longer includes coding agents, the agent shows **Coding is not available here**. To use the agent again, upgrade the plan. A coding session uses [credits](/docs/product/workspaces/credits) for its AI model, by token, and for the time its machine runs. To build a machine, the workspace must have credits. A session that runs on your own API key does not use credits for the model. See [Choose a model](/docs/agents/coding-sessions#choose-a-model). ## Related * [Coding sessions](/docs/agents/coding-sessions) * [Write good tasks](/docs/agents/write-good-tasks) * [Repositories and pull requests](/docs/agents/repositories) * [Automate the coding agent](/docs/agents/automate-coding) * [Machines](/docs/agents/machines) * [Security](/docs/agents/security) --- ## Related pages - [Agents](https://nocodb.com/docs/agents.md): NocoDB agents are AI agents that work in a project: they read and change your tables, documents and code, answer in chat, and run on their own on a trigger. - [AI Agents](https://nocodb.com/docs/agents/ai-agents.md): NocoDB is a no-code AI agent builder for business: AI agents that work on your tables and documents, answer in chat, and run agentic workflows on triggers. - [Create an AI Agent](https://nocodb.com/docs/agents/create-agent.md): How to create an AI agent in NocoDB: describe the job in a setup chat or pick a template, then review and turn it on. NocoAI can also build it. - [AI Agent Use Cases and Templates](https://nocodb.com/docs/agents/use-cases.md): AI agent use cases and examples for sales, customer service, marketing, operations and project management: 13 NocoDB templates to start from. - [Instructions and Files](https://nocodb.com/docs/agents/instructions.md): Write the instructions of a NocoDB agent as a brief for a teammate, mention tables and documents with @, and attach files that the agent reads on every run. - [Tools and Access](https://nocodb.com/docs/agents/tools-and-access.md): Set the AI agent tools of a NocoDB agent: what it can read and change in tables, records, documents and the web, its project role and its managed machine. - [AI Agent Skills](https://nocodb.com/docs/agents/skills.md): Turn on AI agent skills in NocoDB: reusable instructions from the project, workspace or organization catalog that an agent loads when a task needs them. - [Triggers and Results](https://nocodb.com/docs/agents/triggers.md): Build an AI agent workflow in NocoDB with triggers: a schedule, record changes, comments, forms or webhooks. Send the results to chat, email or Slack.