Pre-MVP · Waitlist open
The 100-eyed observability
for AI agents.
Drop-in cost, latency, usage, and error tracking for OpenAI, Anthropic, Gemini, and Mistral. Wrap your SDK client in one line — keep your code unchanged.
SDK shipped(4 providers, 70 tests). Backend ingest scaffolded; hosted ingest + dashboard opening next.
argos + vix = the hundred-eyed giant from Greek mythology.
Four provider SDKs. One observability layer.
OpenAI, Anthropic, Gemini, and Mistral SDKs get the same drop-in wrapper. Aggregate calls across providers under a single recorder.
| Provider | Endpoint | Streaming | Notes |
|---|---|---|---|
| OpenAI | chat.completions.create | Yes | drop-in for openai >= 4.0 |
| OpenAI (responses) | responses.create | Phase B | non-stream observed; streaming planned |
| Anthropic | messages.create | Yes | aggregates message_start + message_delta |
| Mistral | chat.complete + chat.stream | Yes | current production model IDs registered |
| Gemini (legacy) | @google/generative-ai | Yes | getGenerativeModel + generateContentStream |
| Gemini (current) | @google/genai | Yes | ai.models.generateContent + Stream |
5 lines. Drop-in. Done.
Wrap your existing SDK client. Keep all call-site code unchanged. Records ship to the backend ingest endpoint: cost (USD), tokens, latency, error details, and tags per call.
import OpenAI from "openai";
import { wrap, getRecorder } from "@argosvix/sdk";
const client = wrap(new OpenAI(), {
apiKey: process.env.ARGOSVIX_API_KEY,
tags: { service: "support-bot", env: "production" },
});
// Use the wrapped client exactly like the original.
// Every call is recorded transparently.
const completion = await client.chat.completions.create({
model: "gpt-5.5",
messages: [{ role: "user", content: "Hello" }],
});
// Optional: flush before process exit
await getRecorder(client)?.flush();Drop-in patterns for Next.js + OpenAI, Express + Anthropic + streaming, and AWS Lambda + Gemini are shipped with the SDK.
What you get
Cost breakdown
USD cost per call, computed from the current provider pricing table. Filter by tag, model, endpoint, environment.
Latency capture
End-to-end latency per call, including streaming responses. Percentile dashboard on the roadmap (Phase C).
Structured errors
Capture status code, error code, retry-after, and full message. Alert on rate limits before customers notice.
Drop-in proxy
One-line wrap, idempotent re-wrap, transparent return. Streaming, tool calls, all preserved.
Multi-provider
OpenAI, Anthropic, Gemini, Mistral — all in the same recorder. Aggregate across providers per service tag.
TypeScript-first
Full type safety, exactOptionalPropertyTypes-strict, no runtime monkey-patching surprises.
Planned pricing
Final MVP pricing will be set before launch based on real ingestion economics. Join the waitlist for early access and locked-in launch pricing.
Free
Generous free tier (subject to fair use)
- ✓All 4 provider SDKs
- ✓Full SDK access
- ✓7-day retention
- ✓Community support
Starter
Indie & side-project usage
- ✓Everything in Free
- ✓30-day retention
- ✓Custom tags + filters
- ✓Email support
Pro
Small production teams
- ✓Everything in Starter
- ✓90-day retention
- ✓Alert webhooks
- ✓Priority support
Team
Volume-based pricing
- ✓Everything in Pro
- ✓Configurable retention
- ✓SSO + audit log
- ✓Dedicated support
Enterprise (SLA, on-prem, custom retention) — hello@argosvix.com
Be there when the eyes open.
Join the waitlist to get early access + locked-in launch pricing.
No spam. Launch updates only.