Route on your terms.
Choose a fixed model or a routing policy. Balance estimated cost and response time before a request leaves your app.
Choose the model. Set the limits. Route every request through a single gateway, with a clear record of what it costs.
0x14F7CAd1F301827d9CACb6C38751906BdD1196eCKeep model selection, spend limits and fallback behavior in one place. Follow the request from the first token to the final receipt.
Choose a fixed model or a routing policy. Balance estimated cost and response time before a request leaves your app.
Scoped keys are designed to limit how much an agent can spend. The sample workspace lets you test a per-call cap.
Inspect provider costs and routing fees separately. Export a sample receipt to see exactly which fields a call records.
The API design follows the chat-completions request format. Your model choice and policy travel with each call.
Explore the API design ↗from openai import OpenAI client = OpenAI( base_url="https://api.duskr.example/v1", api_key="YOUR_DUSKR_KEY" ) response = client.chat.completions.create( model="duskr/balanced", messages=[{"role": "user", "content": "Explain this contract."}] )
The payment model uses USDG for inference and DUSKR for routing fees and provider bonds. Explore the fee flow below.
Spend on routing.
Bond to serve requests.
The fee model allocates the full inference charge to providers in USDG, plus a 1% routing fee denominated in DUSKR.
No wallet. No funds. A browser-only walkthrough.
Explore routing and itemized costs with fictional providers. No AI calls, wallet transactions or signed proofs are created.
Run a sample request to inspect its selected provider, cost breakdown and receipt.
Fictional models illustrate how a routing catalog could work. Prices are sample per-request costs, not provider quotes.
No models match that search.
DUSKR is an inference routing protocol in development. Developers would send chat-completions requests with a model preference, cost ceiling and fallback policy. This frontend demonstrates that flow with local fixtures.
The sample balanced policy selects Orbit; lowest cost selects Prism; lowest latency selects Orbit. A pinned model overrides the policy. Simulating a failed primary with fallback enabled selects the next available fixture. All checks happen before the sample balances are debited.
The provider receives the USDG inference charge. A separate routing fee equals 1% of that charge, denominated in DUSKR. Half of the fee goes to the treasury, 30% to the reliability reserve and 20% is burned in the fee model.
The demo uses an arbitrary conversion of 1 DUSKR to 0.01 USDG. For a 0.004 USDG request, the routing fee is 0.004 DUSKR: 0.002 treasury, 0.0012 reserve, 0.0008 burn. A production system would need a reliable quote source, slippage limits and enforced settlement. Contract address supplied by the project: 0x14F7CAd1F301827d9CACb6C38751906BdD1196eC. The chain and contract functionality have not been verified here.
A sample receipt includes a request ID, model, policy, cost, fee allocation and creation time. Exported JSON explicitly reports sample mode, unsigned status and no onchain transaction. Cryptographic signatures and proof verification are planned, not implemented.
Browser routing simulation, model search, cost caps, fallback testing, sample balances and JSON exports. State lasts only while this page stays open. Prompts stay in this browser and are not sent to an AI provider.
Accounts, real API keys, model inference, wallet connectivity, USDG settlement, DUSKR issuance, staking/bond contracts and analytics are not connected. The name and ticker are working choices; availability is unverified.