Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/coretracker/agentswarm/llms.txt

Use this file to discover all available pages before exploring further.

AgentSwarm supports two AI providers out of the box: OpenAI Codex and Anthropic Claude. All provider credentials — API keys, base URL overrides, and GitHub tokens — are configured through the Settings UI, not in .env. This keeps secrets out of your filesystem and allows credentials to be rotated without restarting the stack.

​
Where Credentials Live

Provider credentials are stored in the AgentSwarm database under write-only semantics: once saved, raw key values are never returned by the API. The Settings response only tells you whether a credential is configured (openaiApiKeyConfigured: true). To replace a key, simply save a new value; to remove it, use the corresponding clear flag (clearOpenAiApiKey, clearAnthropicApiKey, or clearGithubToken).
All credential endpoints require the settings:edit permission scope. Admin users have this scope by default.

​
OpenAI / Codex Setup

1

Obtain an OpenAI API key

Visit platform.openai.com/api-keys and create a new secret key. Copy it — you will only see it once.
2

Open AgentSwarm Settings and enter your API key

Sign in to AgentSwarm, navigate to Settings, and paste the key into the OpenAI API Key field. Save the form. The openaiApiKeyConfigured flag will flip to true.
3

Set the default Codex model

Choose a codexDefaultModel from the available options:
ValueLabel
gpt-5.4GPT-5.4
o3o3
o4-minio4-mini
o3-minio3-mini
gpt-4.1GPT-4.1
gpt-4oGPT-4o
The default model applied to new tasks is gpt-5.4.
4

Set the default reasoning effort

Choose a codexDefaultEffort value. Codex natively supports three reasoning effort levels:
ValueBehaviour
lowFaster, lower token usage. Best for simple or well-scoped tasks.
mediumBalanced speed and reasoning depth.
highMaximum reasoning depth. Best for complex tasks that require planning across many files.
5

Optionally configure a custom OpenAI-compatible endpoint

If you are using a proxy, a self-hosted model, or an OpenAI-compatible API (such as Azure OpenAI or a local LLM), set openaiBaseUrl to your endpoint. Leave it null to use https://api.openai.com directly.

​
Anthropic / Claude Setup

1

Obtain an Anthropic API key

Visit console.anthropic.com and create an API key under your workspace.
2

Open AgentSwarm Settings and enter your API key

Navigate to Settings in AgentSwarm and paste the key into the Anthropic API Key field. Save the form. The anthropicApiKeyConfigured flag will flip to true.
3

Set the default Claude model

Choose a claudeDefaultModel from the available options:
ValueLabel
claude-opus-4-5Claude Opus 4
claude-sonnet-4-5Claude Sonnet 4.5
claude-sonnet-4Claude Sonnet 4
claude-haiku-3-5Claude Haiku 3.5
The default model applied to new tasks is claude-sonnet-4-5.
4

Set the default reasoning effort

Choose a claudeDefaultEffort value. Claude supports four effort levels — including max, which is Claude-only:
ValueBehaviour
lowMaps to a low thinking budget. Fastest response, least reasoning.
mediumModerate thinking budget.
highLarge thinking budget. Suitable for complex multi-step tasks.
maxLeaves the thinking budget unset, allowing the model to use as much as needed. Claude-only.

​
Per-Task Model Override

When creating or editing a task, you can override the default model by setting modelOverride. This lets you run a specific task at a higher or lower model tier than the system default without changing the global setting. If modelOverride is null or omitted, the task inherits codexDefaultModel or claudeDefaultModel from Settings.

​
Codex Auth Profiles

Codex tasks can authenticate with OpenAI using one of three credential sources, controlled by codexCredentialSource on the task:

auto

AgentSwarm automatically selects the best available credential source. Uses the user’s personal Codex auth JSON if configured; falls back to the global API key.

profile

Uses the requesting user’s personal codexAuthJson credential. Each user can configure their own Codex authentication JSON from their profile settings (UpdateAuthProfileInput.codexAuthJson). Sessions are invalidated when the auth JSON is updated.

global

Uses the shared openaiApiKey configured in Settings. All tasks using this source draw from the same key regardless of which user created the task.
The codexAuthJsonConfigured flag on the user’s auth session indicates whether a personal Codex auth JSON is stored. To clear it, send clearCodexAuthJson: true in an UpdateAuthProfileInput request.

​
GitHub Token

A GitHub personal access token is also configured in Settings (not in .env). It is used for:
  • Importing tasks from GitHub issues and pull requests
  • Posting status updates back to GitHub when syncStatusEnabled is on for a repository
  • GitHub webhook secret validation
Navigate to Settings → Credentials and paste your token into the GitHub Token field. To remove it, send clearGithubToken: true via the credentials API or use the corresponding UI control.

​
Clearing Credentials

To remove a credential without replacing it, use the clear flags available in UpdateCredentialSettingsInput: 
{
  "clearGithubToken": true,
  "clearOpenAiApiKey": true,
  "clearAnthropicApiKey": true
}
Send any combination of these flags in a PATCH /api/settings/credentials request (requires settings:edit). Only the flags you include are acted on; omitting a flag leaves the corresponding credential unchanged.

Build docs developers (and LLMs) love

Get started for freeTalk to us