Skip to main content
Extensions

Agent

Reuse instructions, model settings, and tools for specialized engineering work.

Create a custom agent for recurring work such as code reviews or implementation planning. Give it reusable instructions, a default model, permissions, and the extensions it needs, so you do not have to repeat the same requirements for every task. For a code reviewer, specify which risks to check and ask for a location, impact, evidence, and suggested fix for each finding.

Choose an Agent location

Choose a location based on where you will use the Agent:
LocationUse caseAvailable configuration
LocalWork on the current device with its files and tools.The device’s models, permissions, and extensions.
CloudCollaborate in projects, discussions, and supported Agent Teams, with the required account permissions.Models, Skills, and MCP offered by the cloud form.
Choose local or cloud execution when creating the Agent. You cannot change its location after creation. When managing existing Agents, the list groups them by current device, cloud devices, and other devices.

Configure an agent

Create and manage agents in Settings > Agent, or access them from extension management. Choose Create Agent and select a starting point:
  • Create from blank: enter the name, description, and instructions yourself.
  • Create from template: use the arrow beside a template to preview its Skills and instructions, then click the template to open its configuration. Review and adjust the prefilled settings for your work.
The following steps describe local Agent configuration. For cloud execution, use the models, permissions, and capabilities offered by that form:
  1. Enter or update the name and description. Confirm the name before saving; it cannot currently be changed after creation.
  2. Choose the default model and permissions. These apply to new tasks and collaboration runs. Changing either within a task affects only that task. Local Agents run on the current device. For cloud Agents, use the scope described above.
  3. Write responsibilities, scope, and expected output in Instruction content. If you used a template, review and adjust its instructions.
  4. Select the required Skills, configured MCP servers, and Plugins. Plugins that are not selected are not loaded.
    A template's Skill reference does not mean it is installed locally. If it is unavailable in the current directory, deselect the reference or search the marketplace for the required capability.
  5. To control individual tools, choose Edit under Tool permissions. Allowed tools run without approval, denied tools are blocked, and other tools follow the default permissions. These rules apply only to this agent.
  6. Choose Save Agent. After saving, the agent appears in the list, where you can edit it or select it for a new task.
Example instructions:
Review the supplied code changes for correctness and regression risks.
Read the diff and relevant tests. Report only findings supported by evidence.
Include the location, impact, and suggested fix for each finding.
Do not modify files. Explain what additional information is needed when evidence is missing.

Choose an Agent model

Choose an available model independently for an Agent. When changing a plugin Agent’s model, its plugin continues to manage the supplied instructions and other configuration. A subagent using a custom API-key model needs usable configuration and credentials on its execution device; this does not automatically sync local keys to the cloud or another device.

Create a cloud Agent

To use an Agent in projects or discussions, create it in the cloud:
  1. In Settings → Agent, choose Create Agent → Create from blank.
  2. Set the execution location to Cloud. A blank form defaults to local execution, even when opened from the cloud device list.
  3. Enter the name, description, and instructions. Choose a default model and configure Skills, MCP, and tool permissions as needed. Review the permissions and collaboration scope offered by the cloud form; enterprise policy may restrict available access modes.
  4. Save the Agent and check that it appears enabled in the cloud device list.
  5. Add it when creating a discussion, then select it with @ and send your request. Wait for execution to finish and read its reply in the discussion. See Use in projects and discussions for the detailed steps.
Cloud Agents can collaborate through the options available in Agent Teams. Cloud resources and features still depend on account access and cloud configuration. Enterprise approval policies can also restrict available access modes.

Manage cloud resources

In Settings → Cloud runtime environment, manage Skills, connectors, and memory separately. Select a category and use Sync from local to choose resources. Review the selection and cloud authorization before syncing. Local MCP sync lists only independently configured user-level HTTPS connectors. It excludes stdio, plugin-provided and AppHub-managed connectors, static request headers, and URLs with query parameters.

Create an agent through conversation

Describe an agent's purpose, scope, and expected results in a local task and ask Qoder to draft its configuration. For example: “Create a code reviewer that checks regression risks without modifying files.”
  1. Review the draft's name, description, and instructions in the conversation.
  2. Select Continue configuration to open the configuration form.
  3. Check the model, permissions, and extensions, adjust them as needed, and select Save Agent.
  4. After saving succeeds, select the agent from the executor menu in a new task.
A generated draft is not a saved agent. You can select the agent in a new task only after saving it successfully in the configuration form. Conversation-based creation is available in local tasks. Remote SSH and WSL remote tasks are not supported.

View Agents on other devices

After connecting a device, use Other devices to view its Agents and online status. The directory is read-only. Confirm the target device and workspace authorization before running a task.

Use an agent

Select an agent in a new task to have it handle the work from the start. During an existing local conversation with Qoder, you can also delegate part of the work and ask for the result to be returned.

Start a task

After creating and saving a local Agent, select it for a new task:
  1. Click New Task in the left sidebar to open the new task page.
  2. At the top of the page, click the Qoder name or the arrow next to it to open the executor picker.
  3. Select your agent from the Custom Agents section. Check that Qoder changes to the selected agent's name, then enter and send your request.
Open the Qoder menu on the new task page and select a custom agent
To create an agent, choose Customize Agents at the bottom of the menu; for an existing agent, you can also start a task from Task at the top of its editor.

Delegate through conversation

When a local conversation with Qoder reaches a point that needs a specialist review, delegate that part of the work without starting over with the context. First check that the agent is enabled and available in Settings > Agent. Name the agent, identify the material, and describe the result you need. For example:
Give the API change proposal above to my code-review agent. Check compatibility risks and suggest tests.
Analyze only; do not modify files. Follow the task to completion and bring the result back here. Tell me if a decision needs my input.
Qoder creates a task for the agent. When you request follow-up, results and pending questions return to the current conversation; you can also open the delegated task to inspect its progress. Check the returned findings against your request. If the response only confirms task creation, the work is still pending.

Use in projects and discussions

Projects and discussions require a business plan (Teams or Enterprise). Not yet available on individual plans.
For a local Agent, enable Use in Issues and discussions in its settings and keep the current device online. Work on a project issue
  1. Open an issue in a project and start a discussion. Confirm that its title, description, and issue association are carried over.
  2. Select your agent through Agent, then choose Create.
  3. Invoke the agent as described below. Record any follow-up work in the linked issue.
Invoke an agent in a discussion
  1. Select the agent through Agent when creating a discussion, or use Share > Add Agent / Agent Team in an existing discussion.
  2. Type @, open the Agent tab, select your agent, and send your request.
  3. Wait for execution to finish and read the reply in the discussion. For local Agents, you can also choose Open execution chat to inspect progress and status.
You can only invoke agents you own with @; you cannot invoke agents added by other members.

Update an agent

In Settings > Agent, use the edit button, update the description, instructions, or execution settings, then choose Save Agent. To stop using an agent temporarily, turn off its enable switch in the list or Available for assignment in the editor. This interrupts unfinished tasks and stops new tasks; historical records remain. Cloud Agent settings support renaming. After saving, verify the name and availability in the list. Local Agent naming rules remain governed by their configuration page.

Manage archived Agents

Check the object in the archived list before using an available restore or delete action. Restoring a cloud Agent requires the relevant permission; verify its availability afterward. Read the impact notice before permanent deletion, which is different from archiving.