Skip to main content

​
Bifrost

Bifrost is an LLM proxy that routes requests to multiple backends — Anthropic, OpenAI, Bedrock, Gemini, Ollama, and more — with routing rules, prompt cache optimisation, semantic caching, fallback chains, and cost tracking. WednesdayAI communicates with Bifrost as a single provider (bifrost) using an OpenAI-compatible API. Bifrost handles all model-specific logic internally.
Bifrost is self-hosted. You must run a Bifrost instance (locally or on your own infrastructure) before configuring WednesdayAI to use it.

​
Prerequisites

  • A running Bifrost instance. See the Bifrost documentation for setup.
  • Optionally, a Bifrost virtual key for authentication.

​
Quick setup

​
1. Start Bifrost

Follow the Bifrost setup guide to start your instance and note its base URL.

​
2. Configure WednesdayAI

openclaw auth
# Choose: Proxy / Gateway → Bifrost
Enter your base URL, optional virtual key, and the model aliases you want available (e.g. medium, fast, anthropic/claude-sonnet-4-6).

​
3. Manual config snippet

{
  models: {
    providers: {
      bifrost: {
        baseUrl: "http://localhost:8080", // your Bifrost instance URL
        apiKey: "BIFROST_API_KEY",        // optional virtual key; omit if not used
      },
    },
  },
  agents: {
    defaults: {
      model: { primary: "bifrost/medium" }, // routing alias defined in Bifrost
    },
  },
}
apiKey is the name of the environment variable, not the key value itself. Keep secrets in .env, not in committed config files.

​
Model aliases vs direct model IDs

You can use either:
  • Routing aliases (medium, fast) — Bifrost maps these to real models via its routing rules.
  • Direct model IDs (anthropic/claude-sonnet-4-6) — bypasses routing and targets that model directly.
Bifrost decides which backend and model to use. WednesdayAI does not influence routing.

​
Provider settings

SettingValue
Provider IDbifrost
APIOpenAI-compatible
Auth env varBIFROST_API_KEY (optional)
Default timeout600 seconds

​
Usage dashboard

When Bifrost is configured to forward response headers, WednesdayAI records both the configured alias and the actual backend provider/model per call:
  • byProvider shows actual provider spend (e.g. anthropic: $0.42)
  • byConfiguredModel shows spend by alias (e.g. medium: $0.42)

​
Caching

Bifrost’s prompt cache optimisation and semantic cache run on the Bifrost side. WednesdayAI does not send cache_control markers to Bifrost — Bifrost’s CacheOptimizer analyses request structure autonomously.

​
Verification

openclaw models status --probe-provider bifrost

​
Troubleshooting

Verify your Bifrost instance is running and the baseUrl in your config matches its address and port.
This is an upstream provider defect, not a WednesdayAI or Bifrost bug. Some backends (observed on Cloudflare- and Parasail-hosted vLLM endpoints) emit finish_reason=tool_calls without the accompanying tool-call chunks.Fix: Exclude the offending upstreams at the Bifrost routing layer — pin routes to compliant backends (e.g. SiliconFlow) and exclude non-compliant ones (e.g. Cloudflare).
If you configured a virtual key, verify BIFROST_API_KEY is set correctly in ~/.openclaw/.env.