What ships, what's next, and what to integrate against.
The technical view of every Vraelis surface. Status, version, what it depends on, and how the API exposes it. For the consumer walkthrough, see the home page.
Surfaces
OpenAI-compatible chat, plus the parts that aren't OpenAI.
Drop-in replacement for the OpenAI SDK. Chat, streaming, tool use, function calling — same shape. Beyond that: project-scoped memory, file uploads, voice in / out, and an MCP runtime that picks up any server you point at it.
/v1/chat/completions/v1/projects/v1/projects/{id}/memory/v1/projects/{id}/files/v1/voice/transcribe/v1/voice/speak/v1/mcp/servers/v1/webhooks/v1/lens/renderTypeScript / Python, identical surface
import Vraelis from "@vraelis/sdk";
const sx = new Vraelis({ apiKey: process.env.VRAELIS_API_KEY });
const stream = await sx.chat.completions.create({
model: "vraelis-1",
project: "proj_5g7",
stream: true,
messages: [{ role: "user", content: "summarize my open PRs" }],
tools: [{ type: "mcp", server: "github" }],
});
for await (const chunk of stream) {
process.stdout.write(chunk.delta.text ?? "");
}First-party connectors and any MCP server.
First-party connectors handle auth + rate limiting for the tools we ship. Anything else, point at an MCP server and the runtime picks it up automatically.
Completion finished, tool call invoked, key spend exceeded, project usage cap hit. HMAC-signed delivery, configurable retry, per-project endpoints.
What tool, what input, what output, what time. Replayable from the audit page. Most write actions are reversible for 24 hours.
Per-project keys with spend caps, request rate limits, expiry dates. Real-time tail of every request the model is handling for you.
Build on it before it ships everywhere.
The Platform is alpha. API surface is stable enough to integrate against; behavior may shift before v1. Get a key and start.