Prove your AI behaved — not just that it was fast.

Passive, never autonomous

monsys.ai doesn't run prompts, doesn't take actions, never blocks inline. It's an observability layer — after-the-fact evidence, not a control plane.

PII redacted at source

Belgian IBAN, RRN, BTW, KBO, emails and phones are detected with checksum validation before storage and replaced with a hash token. Raw content never reaches the hub.

Ed25519-signed evidence packs

One click per period → gzipped tarball with manifest + signature. A standalone Python script verifies offline, with no monsys account. Built for auditors and regulators.

Python

from monsys_ai import Tracer
tracer = Tracer()

with tracer.trace("rag.chat") as t:
  with t.span("openai.chat",
              provider="openai",
              model="gpt-4o") as s:
    s.prompt = user_msg
    r = openai.chat.completions.create(...)
    s.completion = r.choices[0].message.content
    s.input_tokens  = r.usage.prompt_tokens
    s.output_tokens = r.usage.completion_tokens

Node / TypeScript

import { Tracer } from "./monsys-ai";
const tracer = new Tracer();

await tracer.trace("rag.chat", async (t) => {
  await t.span("openai.chat",
    { provider: "openai", model: "gpt-4o" },
    async (s) => {
      const r = await openai.chat.completions.create({...});
      s.record({
        prompt: userMsg,
        completion: r.choices[0].message.content!,
        inputTokens:  r.usage!.prompt_tokens,
        outputTokens: r.usage!.completion_tokens,
      });
    });
});

Go

tracer, _ := monsysai.New(monsysai.Options{})

err := tracer.Trace(ctx, "rag.chat", func(t *monsysai.Trace) error {
  return t.Span(ctx, "openai.chat",
    monsysai.SpanOpts{Provider: "openai", Model: "gpt-4o"},
    func(s *monsysai.Span) error {
      s.Prompt = userMsg
      // call your LLM...
      s.Completion = resp
      in, out := inTok, outTok
      s.InputTokens  = &in
      s.OutputTokens = &out
      return nil
    })
})