For developers
One endpoint. Every model you need.
POST a request, get back a structured run — model selection, tool calls, and retries handled for you. SDKs for Python, Node, and Go, or call the REST API directly.
curl https://api.apexvyra.example/v1/runs \
-H "Authorization: Bearer $APEXVYRA_KEY" \
-H "Content-Type: application/json" \
-d '{
"agent": "refund-triage",
"input": "Refund order #4821"
}'from apexvyra import APEXVYRA
client = APEXVYRA(api_key="APEXVYRA_KEY")
run = client.runs.create(
agent="refund-triage",
input="Refund order #4821",
)
print(run.output)import APEXVYRA from "apexvyra";
const client = new APEXVYRA({ apiKey: process.env.APEXVYRA_KEY });
const run = await client.runs.create({
agent: "refund-triage",
input: "Refund order #4821",
});
client := apexvyra.NewClient(os.Getenv("APEXVYRA_KEY"))
run, err := client.Runs.Create(ctx, &apexvyra.RunParams{
Agent: "refund-triage",
Input: "Refund order #4821",
}){
"id": "run_8841",
"status": "succeeded",
"model": "balanced",
"output": "Refund of $42.00 issued for order #4821.",
"latency_ms": 428,
"cost_usd": 0.0021
}One request, one run
Every run is a single request — and a full trace
You don't manage a queue, a retry policy, or a separate logging call. One request returns the result and everything that happened while producing it: which model ran, which tools it called, and what each step cost.
{
"id": "run_8841",
"trace": [
{ "step": "route", "model": "balanced" },
{ "step": "tool_call", "name": "lookup_order" },
{ "step": "guardrail", "check": "pii_redaction" }
]
}Quickstart
From API key to first run in four steps
-
1
Get your API key
Create a free workspace and generate a key from Settings → API Keys.
-
2
Install the SDK
pip install apexvyra,npm install apexvyra, orgo get apexvyra— or skip this and call the REST API directly. -
3
Make your first request
Send an
inputstring to an agent. You'll get a run object back with the result. -
4
Add routing, tools, and guardrails
Layer in a routing policy, tool definitions, and guardrails as your agent grows — no changes to the calling code.
Built for the API
Handled for you, so your integration code stays small
Streaming responses
Token-by-token output over Server-Sent Events, or wait for the full run.
Structured outputs
Constrain a run's output to a JSON schema you define.
Function calling
Declare tools once; every routed model gets a compatible calling format.
Automatic retries
Transient provider failures retry with backoff before they reach you.
Rate limit handling
Requests queue and smooth out automatically as you approach a limit.
Idempotency keys
Retry a request safely — pass the same key and get the same run back.
Changelog
Recently shipped
-
Streaming tool calls
Tool calls now stream incrementally instead of arriving as one block at the end of a run.
-
Go SDK
A first-party Go client, matching the existing Python and Node SDKs.
-
Idempotency keys
Pass an idempotency key to
/v1/runsto make retries safe.
Read the docs, then ship something.
Full API reference, SDK guides, and example agents.