AI Gateway
Build with leading models through a common API. Choose your model, route eligible requests with DOSRouter, and keep usage and access controls in one place.
POST /v1/chat/completionsCapabilities and request controls depend on the selected model, endpoint, and configuration.
Capabilities
Start with a common API, then add the routing, request policies, and usage visibility your application needs.
Classify request signals and select an eligible route. Use automatic routing or select a model explicitly when you need a specific capability.
Route selection follows the configured model pool and availability. Model capabilities still determine which requests are supported.
Review requests, input and output tokens, costs, and model performance from your dashboard.
Understand usage by model and time period, then use those measurements to guide your next integration decision.
Apply deterministic checks for sensitive data and secrets in supported inference requests. Configured policies can flag, mask, or block matching input.
Request policies can tighten the server baseline. Coverage depends on the supported detectors and the request path.
Manage API keys, control access to your resources, and review organization activity.
Provider-specific processing and retention conditions still apply. Review model and provider details before sending sensitive data.
Reuse eligible work instead of processing the same context or request from scratch.
Prompt caching reuses shared context on supported providers. Eligibility, retention, and cache rates vary by model.
Response caching reuses a stored response for an eligible matching request. Other requests continue to the model.
Reduce repeated context through deduplication, whitespace trimming, and JSON compaction on supported requests.
Compression runs only when the request meets its threshold and the resulting input is smaller.
Use the OpenAI-compatible API or the supported Anthropic Messages endpoint. Check the selected model for streaming, vision, and tool support.
Create a key in the console and keep it in your server environment.
Point your compatible client at https://api.dos.ai/v1.
Use a model ID from the catalog. Review usage and cost in the console.
from openai import OpenAI
import os
client = OpenAI(
base_url="https://api.dos.ai/v1",
api_key=os.environ["DOS_API_KEY"],
)
response = client.chat.completions.create(
model="dos",
messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)Browse currently listed models and inspect capabilities before integrating. Model membership and pricing come from the same catalog used by the platform.
60models in the current catalog
OpenAI-compatible clients can use the DOS base URL and API key. The Anthropic Messages endpoint is also supported. Model-specific request fields and capabilities still apply; see the API documentation for details.
No. Context length, modalities, tool use, streaming, and pricing vary. Check the model detail page and select a model that supports your workload.
Use appropriate request policies and review provider processing and retention conditions. Safeguard checks supported patterns; it does not replace your application access controls or data policy.
Charges follow the model and its published unit, such as tokens, images, or audio duration. Input, output, and supported cache rates can differ. Agent hosting is a separate service.
Start with an API key, or explore DOSClaw if you want a hosted agent with knowledge and integrations.