Every AI_SDK_* Django setting, with its default and purpose. Most are optional: the SDK works with sensible defaults out of the box.

Settings are read via getattr(settings, ...) at call time (cached where noted), so none of them require a restart of the dev server beyond Django’s normal settings reload.

Agent Registration and Runtime Agents

Setting	Default	Purpose
AI_SDK_AGENTS	[]	Dotted module paths to import so their Agent subclasses auto-register. See the Agents guide and Agent Registry.
AI_SDK_RUNTIME_AGENT_BASES	[]	Base classes a DB-configured runtime agent can be built on, as dotted paths. See Runtime-Configured Agents.
AI_SDK_RUNTIME_AGENT_TOOLS	{}	Tools selectable in runtime agent configuration: {"key": "path.to.provider"}.

Models

Setting	Default	Purpose
AI_SDK_DEFAULT_MODEL	"gpt-4o-mini"	Fallback model identifier for llm_generator() when no model is otherwise configured.
AI_SDK_EXTRACTION_MODEL	None	Model used by the document-extraction generator (agents/utils.py). Falls back to AI_SDK_DEFAULT_MODEL when unset.

RAG and Vector Stores

Setting	Default	Purpose
AI_SDK_VECTOR_STORE_PATH	None	Base directory for persistent vector stores. Storage configs build per-backend, per-memory paths as {path}/{backend}/{memory_id}; unset means in-memory stores. See RAG Variants.
AI_SDK_VECTOR_STORE_URL	None	Qdrant server URL. When set, Qdrant storage uses backend="server" with collection index memory_{memory_id} (or default).

Files and Uploads

Setting	Default	Purpose
AI_SDK_ALLOWED_FILES	{}	Extra file-extension → MIME mapping used as a fallback when magic-byte detection fails (files/processors.py).
AI_SDK_MAX_UPLOAD_SIZE	10 MB	Upload size ceiling surfaced to the frontend by get_upload_settings().
AI_SDK_MEMORY_FILE_PIPELINE	None	Dotted path (or list of paths) to a zero-argument callable returning a FilePipeline: the default pipeline for uploads without agent context. The first pipeline whose accepts(file) matches is used. See Files.
AI_SDK_FILE_PIPELINE_TIMEOUT	900	Seconds before a background document pipeline is failed. django_tasks has no native timeout, so this guards hung or wedged upload processing.

MCP Servers

Setting	Default	Purpose
AI_SDK_MCP_DISCOVERY_CACHE_TTL	3600	Seconds OAuth discovery results are cached in-process.
AI_SDK_MCP_DISCOVERY_TIMEOUT	10	Seconds for a single OAuth discovery request.
AI_SDK_MCP_ALLOWED_ISSUER_DOMAINS	None	Allow-list of OAuth issuer domains (e.g. ["accounts.notion.com"]). None allows any issuer; [] rejects all. Defense-in-depth against compromised MCP servers.
AI_SDK_MCP_CLIENT_NAME	"MCP OAuth Client"	Client name registered with OAuth servers during the authorization flow.
AI_SDK_MCP_SERVER_LIST_CACHE_TTL	30	Seconds the integrations registry caches its server list.
AI_SDK_MCP_REFRESH_THRESHOLD_MINUTES	10	Refresh OAuth tokens proactively when they expire within this many minutes.
AI_SDK_MCP_OAUTH_SUCCESS_URL	"/"	Redirect target after a successful MCP OAuth flow.

Integrations

Setting	Default	Purpose
AI_SDK_INTEGRATIONS	{}	Per-integration configuration dict, keyed by registry name, in the shape of DATABASES/CACHES. Keys are upper-cased on read. Secrets are referenced inline, e.g. {"github": {"TOKEN": env("GITHUB_MCP_TOKEN")}}: there is no derived AI_SDK_GITHUB_TOKEN variable, because GitHub Actions already injects a conflicting GITHUB_TOKEN. See the Integrations guide.
AI_SDK_INTEGRATION_TIMEOUT	3	Seconds per integration tool call.
AI_SDK_INTEGRATION_CACHE_TTL	900	Seconds tool lists and statuses are cached.
AI_SDK_INTEGRATION_CB_COOLDOWN	60	Seconds a failed integration stays open-circuited before retry.

Permissions

Setting	Default	Purpose
AI_SDK_PERMISSIONS	{}	Per-domain overrides: {"memory": ["path.to.PermissionClass"], ...}. Domains: agent, thread, memory, integrations. See Permissions.

Workflows

Setting	Default	Purpose
AI_SDK_WORKFLOW_ACTIONS	{}	Workflow action registry: {"console_log": "path.to.ConsoleLogAction"}. See Workflows.

Streaming and Titles

Setting	Default	Purpose
AI_SDK_STREAM_CORS_ORIGIN	None	Allowed origin for SSE stream responses (used alongside django-cors-headers).
AI_SDK_SUGGESTION_TIMEOUT	5.0	Seconds before suggestion generation is abandoned.
AI_SDK_TITLE_SANITY_LIMIT	80	Character ceiling for auto-generated thread titles.

Logging

Setting	Default	Purpose
AI_SDK_ENABLE_LOGS	False	When True, adds a DEBUG-level loguru handler to stderr. File logging to logs/django_ai_sdk.log runs regardless.

Provider Credentials

Not AI_SDK_* prefixed, but part of the same surface: the generator and RAG libraries read them directly:

Setting	Purpose
OPENAI_API_KEY	Provider API key (often injected as an environment variable).
OPENAI_API_URL	Optional custom API base URL (rags pipelines and agents/utils.py read it).