2.3.84 Satellite Bifrost

Bifrost

Handle: bifrost
URL: http://localhost:34861/workspace/logs

Bifrost is a high-performance AI gateway that exposes an OpenAI-compatible API over local and remote model providers. Harbor runs it with persistent SQLite-backed configuration and can bootstrap Harbor backends such as Ollama, llama.cpp, Docker Model Runner, MLX, and oMLX as Bifrost providers.

Starting

harbor pull bifrost
harbor up bifrost --open

The dashboard starts with no required credentials. Configure cloud providers in the UI, or start Bifrost together with Harbor backends to auto-register them:

harbor up bifrost ollama
harbor up bifrost llamacpp
harbor up bifrost ollama llamacpp

Bifrost's OpenAI-compatible endpoint is available at:

http://localhost:34861/v1

From inside the Harbor network, use:

http://bifrost:8080/v1

Harbor integrations

Ollama

When bifrost and ollama run together, Harbor starts a one-shot bootstrap container that registers an ollama provider pointing at ${HARBOR_OLLAMA_INTERNAL_URL}.

Use Bifrost models with the provider prefix, for example:

curl http://localhost:34861/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer sk-bifrost' \
  -d '{
    "model": "ollama/qwen3.5:4b",
    "messages": [{"role": "user", "content": "hello"}]
  }'

llama.cpp

When bifrost and llamacpp run together, Harbor registers a custom OpenAI-compatible provider named llamacpp pointing at http://llamacpp:8080.

harbor up llamacpp bifrost

By default, Harbor discovers model IDs from http://llamacpp:8080/v1/models during bootstrap and registers them with Bifrost. Then call models through Bifrost using the llamacpp/ provider prefix, for example llamacpp/Qwen3-0.6B-Q4_K_M.gguf.

Docker Model Runner

When bifrost and dmr run together, Harbor registers a custom OpenAI-compatible provider pointing at http://dmr:8080 using HARBOR_BIFROST_DMR_*.

harbor up dmr bifrost

Call models through Bifrost with the dmr/ prefix, for example dmr/ai/smollm2.

MLX

When bifrost and mlx run together, Harbor registers MLX at http://mlx:8080 using HARBOR_BIFROST_MLX_*.

harbor up mlx bifrost

Call models through Bifrost with the mlx/ prefix, for example mlx/mlx-community/Qwen3.5-4B-4bit.

oMLX

When bifrost and omlx run together, Harbor registers oMLX at http://omlx:8080 using HARBOR_BIFROST_OMLX_*.

harbor up omlx bifrost

Call models through Bifrost with the omlx/ prefix, for example omlx/Qwen3.5-4B-4bit.

Open WebUI

harbor up bifrost webui

This adds Bifrost to Open WebUI as an OpenAI-compatible backend.

Harbor Boost

harbor up bifrost boost

This registers Bifrost as a named downstream API for Harbor Boost.

Configuration

Environment Variables

Following options can be set via harbor config:

HARBOR_BIFROST_HOST_PORT              # Host port for the Bifrost UI/API
HARBOR_BIFROST_IMAGE                  # Container image
HARBOR_BIFROST_VERSION                # Container image tag
HARBOR_BIFROST_WORKSPACE              # Persistent Harbor workspace
HARBOR_BIFROST_APP_PORT               # Internal Bifrost HTTP port
HARBOR_BIFROST_INTERNAL_URL           # Internal Harbor URL for other services
HARBOR_BIFROST_OPEN_URL               # Browser URL opened by `harbor open bifrost`
HARBOR_BIFROST_API_KEY                # Placeholder bearer token used by Harbor clients
HARBOR_BIFROST_LOG_LEVEL              # Bifrost log level
HARBOR_BIFROST_LOG_STYLE              # Bifrost log format
HARBOR_BIFROST_BOOTSTRAP_IMAGE        # Provider bootstrap helper image
HARBOR_BIFROST_BOOTSTRAP_VERSION      # Provider bootstrap helper image tag
HARBOR_BIFROST_OLLAMA_MODELS          # Models allowlist for the auto-registered Ollama key
HARBOR_BIFROST_LLAMACPP_BASE_URL      # Internal llama.cpp base URL registered in Bifrost
HARBOR_BIFROST_LLAMACPP_API_KEY       # API key sent from Bifrost to llama.cpp
HARBOR_BIFROST_LLAMACPP_MODELS        # Models allowlist for llama.cpp; `*` auto-discovers /v1/models
HARBOR_BIFROST_DMR_BASE_URL           # Internal Docker Model Runner base URL registered in Bifrost
HARBOR_BIFROST_DMR_API_KEY            # API key sent from Bifrost to DMR
HARBOR_BIFROST_DMR_MODELS             # Models allowlist for DMR; `*` auto-discovers /v1/models
HARBOR_BIFROST_MLX_BASE_URL           # Internal MLX base URL registered in Bifrost
HARBOR_BIFROST_MLX_API_KEY            # API key sent from Bifrost to MLX
HARBOR_BIFROST_MLX_MODELS             # Models allowlist for MLX; `*` auto-discovers /v1/models
HARBOR_BIFROST_OMLX_BASE_URL          # Internal oMLX base URL registered in Bifrost
HARBOR_BIFROST_OMLX_API_KEY           # API key sent from Bifrost to oMLX
HARBOR_BIFROST_OMLX_MODELS            # Models allowlist for oMLX; `*` auto-discovers /v1/models

Harbor also forwards shared provider keys such as HARBOR_OPENAI_KEY, HARBOR_ANTHROPIC_KEY, HARBOR_MISTRAL_KEY, HARBOR_GROQ_KEY, and HARBOR_COHERE_KEY into the Bifrost container using the environment variable names expected by Bifrost examples.

Volumes

${HARBOR_BIFROST_WORKSPACE}:/app/data

Bifrost stores its SQLite configuration database, request logs database, and optional config.json under /app/data.

Troubleshooting

harbor logs bifrost

Check the health endpoint:

curl http://localhost:34861/health

If a backend was not registered, inspect the bootstrap sidecar logs:

docker logs harbor.bifrost-ollama-bootstrap
docker logs harbor.bifrost-llamacpp-bootstrap
docker logs harbor.bifrost-dmr-bootstrap
docker logs harbor.bifrost-mlx-bootstrap
docker logs harbor.bifrost-omlx-bootstrap

If you already configured Bifrost manually, the bootstrap sidecars update the Harbor-managed provider and key IDs rather than replacing unrelated providers.

Links

• Official Documentation
• GitHub Repository
• Provider Configuration