> ## Documentation Index
> Fetch the complete documentation index at: https://felan.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Agent
> Configure the models, extensions, skills, and subagents available to your Felan team.
Open **Agent** from the dashboard sidebar to configure how Felan works for the selected team. The page has four tabs:
| Tab | What it controls |
| -------------- | --------------------------------------------------------------------------------------- |
| **Models** | The root agent's default provider and model, plus team-owned provider credentials |
| **Extensions** | Executable capabilities and tools composed into the hosted agent runtime |
| **Skills** | Reusable instructions and supporting resources imported from public GitHub repositories |
| **Subagents** | Built-in and team-created specialists to which the main agent can delegate work |
Owners and admins can change Agent configuration. Team members can view extensions, skills, and subagent metadata, but cannot change them. Configuration changes apply to new sessions; they do not reconfigure an active session.
## Models
The default provider and model run the root agent that communicates with your team and coordinates delegated work. You can select a provider supplied by the Felan platform or configure a team provider override.
Team providers support the authentication methods shown in the provider catalog:
* **API key** stores a provider key for the team.
* **OAuth** uses a hosted device-code flow for supported subscription-backed providers, including GitHub Copilot and ChatGPT Plus or Pro.
Provider credentials are encrypted, remain server-side, and are never shown in full after saving. The model list contains the models supported for the selected provider by Felan's runtime catalog.
A subagent can inherit this default, select a portable model tier, or use an exact configured model. Changing a subagent's model does not change the root agent's default.
## Extensions
Extensions are independently published runtime packages that add tools or agent behavior. Examples include browser automation, subagent delegation, task tracking, public web access, document conversion, and access to configured MCP servers.
The Extensions tab reports one of these states for each package:
| State | Meaning |
| --------------- | ------------------------------------------------------------------------- |
| **Available** | The extension can be enabled or disabled for new sessions |
| **Required** | Felan needs the extension for a hosted workflow, so it cannot be disabled |
| **Unavailable** | The team does not currently meet an integration or plan requirement |
Some extensions depend on another extension. Felan validates those dependencies when you save. For example, prewalk planning requires task tracking. The MCP extension is available only when the team has an active MCP configuration, and the memory extension depends on team memory being enabled.
Disabling an extension removes its capability from new sessions. It does not delete provider data or change an already-running session.
## Skills
A skill is a reusable set of instructions, examples, and supporting files that teaches the agent how to perform a workflow. Skills guide model behavior; extensions provide executable tools. A workflow may use both.
Team skills are imported from a public GitHub repository:
Provide a public GitHub repository and a branch or tag.
Felan finds valid `SKILL.md` files and displays their names, descriptions, and paths.
Choose the skills to add to the team catalog.
Felan pins the selected skills to the exact commit that was scanned.
Imported skills are available to the main agent and team subagents in new sessions. The complete skill directory is installed, so a skill can include references and other supporting files next to `SKILL.md`.
Use **Refresh** to rescan the saved branch or tag and pin the skill to its latest resolved commit. Use **Remove** to remove it from future sessions. Neither action changes the source repository.
Skill imports accept public GitHub repositories only. Private repositories, other Git hosts, local paths, malformed skills, and oversized `SKILL.md` files are rejected.
## Subagents
Subagents are specialized agents that the main Felan agent can delegate focused work to. They run as child sessions, return their result to the parent, and share the root session's workspace. The main agent remains responsible for coordinating the final outcome and for source-control publication.
Felan currently provides these visible built-ins:
| Built-in | Intended work |
| ----------- | -------------------------------------------------------------------- |
| **Explore** | Read-only repository and technical exploration |
| **General** | Investigation, debugging, implementation, and general technical work |
| **Review** | Read-only code and pull-request review through a specified lens |
Owners and admins can customize a built-in's name, delegation description, prompt, and model. **Reset to default** removes the team's override and restores the Felan-provided definition; it does not delete the built-in.
You can also create up to 50 reusable custom subagents. Each custom subagent has:
* A unique name
* A description that helps the main agent decide when to delegate to it
* Prompt and instructions for completing the delegated work
* A model selection
Custom subagents can be edited or deleted. Built-ins can be customized and reset, but not deleted. Team-authored instructions cannot override platform safety, authorization, tenant-isolation, credential, or execution boundaries.
### Choose a subagent model
| Selection | Behavior |
| ----------------------------------------- | -------------------------------------------------------------------------------- |
| **Inherit team default** | Uses the default provider and model selected in Models |
| **Extra high, High, Medium, or Low tier** | Resolves to an authenticated model in that capability tier when the child starts |
| **Exact `provider/model`** | Always requests that configured, supported model |
Portable tiers let the same subagent configuration work when a team's available providers change. Resolution happens when the subagent starts, and normal availability, credential, entitlement, and usage checks still apply. A tier is a model-capability choice; it is separate from the model's reasoning or thinking level.
Imported team skills are available to built-in and custom subagents. Enabled extensions are composed into subagent runtimes where the extension supports that role.
## What changes affect
Agent configuration is team-scoped. A saved change affects sessions started after the save:
* Existing root and child sessions keep the configuration with which they started.
* A new session receives the current default model, enabled extensions, imported skills, and subagent catalog.
* Removing or resetting configuration does not alter previous session transcripts or results.
## API access
Team API keys can manage the same team skill catalog and customer-defined custom subagents under `/api/v1/teams/{teamSlug}`. Use the **Skills** API to scan a public GitHub repository, import selected `SKILL.md` paths at the scanned commit, refresh a saved source, or remove a skill. Use the **Subagents** API to list, read, create, update, or delete definitions with the supported `name`, `description`, `instructions`, and `model` fields.
The API exposes custom subagents only. It cannot read or change built-ins, the root agent, provider credentials, per-subagent skill or extension selections, or execution limits. Imported team skills remain shared by the root agent and all subagents. Removing a skill removes it from future sessions without leaving per-subagent references.
See the **API Reference** tab for endpoint bodies, responses, and error codes. Configuration writes require a team API key for the team named by `teamSlug`.
See how the main agent coordinates sessions and delegated work.
Start work and inspect delegated subagent activity.