Configuration
Key environment variables
| Variable | Values | Description |
|---|---|---|
DEPTHFUSION_MODE |
local · vps-cpu · vps-gpu · mac-mlx |
Deployment mode. Controls which embedding backend and server transport are used. Defaults to local. |
DEPTHFUSION_REST_API |
true / http://<host>:<port> |
Set true to start the REST API on 127.0.0.1:7300. Set to a URL to point the MCP server at a remote instance. |
DEPTHFUSION_DISTILLATION_BACKEND |
bm25 · openai · mlx |
Retrieval/embedding backend. bm25 requires no API key; openai requires OPENAI_API_KEY; mlx requires Apple Silicon. |
DEPTHFUSION_MCP_TOKEN |
any string | Bearer token required on all /sse and /messages requests when DEPTHFUSION_REST_API is active. Fail-closed — set this before exposing the server beyond loopback. |
DEPTHFUSION_JWKS_URI / DEPTHFUSION_OIDC_ISSUER / DEPTHFUSION_OIDC_AUDIENCE |
URL / string / string | OIDC identity provider config for real, capability-bearing principals (as opposed to the capability-limited legacy token path). See docs/mcp-client-setup.md § OIDC Device-Code Login. |
Profiles
DepthFusion ships three named profiles that set a sensible baseline:
| Profile | DEPTHFUSION_PROFILE |
Use case |
|---|---|---|
minimal |
minimal |
Laptop / CI — BM25 only, no AI passes |
standard |
standard (default) |
Self-hosted VPS — BM25 + session tracking |
research |
research |
Full feature set — requires embedding backend + API key |
# Example: standard profile with cognitive scoring override
DEPTHFUSION_PROFILE=standard
DEPTHFUSION_COGNITIVE_SCORING=true
Individual env vars always win over the profile baseline.
Inspecting active config
# Via MCP tool
depthfusion_status
# Or check the active profile flags
depthfusion_describe_capabilities
Full configuration reference → configuration-profiles.md