Viktor.comAn AI coworker that actually does the work
Observyze
Real-time LLM observability & guardrails — 100% free
3 followers
Real-time LLM observability & guardrails — 100% free
3 followers
Observyze is back in Early Access! Most LLM logging tools add 200ms+ overhead. Observyze’s proxy-first architecture adds <100ms while capturing high-fidelity traces for prompts, tool calls, and agent workflows.
Features:
• Token-level cost attribution
• Redis-backed circuit breakers & guardrails
• Prompt A/B testing & regression detection
• In-memory PII scrubbing
2 lines of code. No infra.
🚀 100% Free: Claim 90 days of full Pro access with no credit card needed!
This is the 2nd launch from Observyze. View more
Observyze
Launching today
Since our beta, we've rebuilt Observyze for production scale. What's new in Early Access:
⚡ Sub-10ms Ingestion: Non-blocking telemetry proxy with zero latency drag.
🔒 Client-Side PII Scrubbing: In-memory masking before network transmission.
🛑 Active Circuit Breakers: Halts hallucinations before reaching users.
💰 40+ Model Cost Tracking: Exact token-level spend attribution.
🎉 100% Free: Claim 90 days of full Pro tier (100k traces/mo) with no credit card needed!
Free Options
Launch Team / Built With
Hey Product Hunt community! 👋
I'm Rahul, the maker behind Observyze.
When building AI applications and autonomous agents, we kept running into the exact same three friction points:
1.Passive Logging is Too Slow: Traditional observability tools add 150ms–200ms+ overhead per LLM call, killing user-facing response times.
2.Post-Mortem Hallucination Detection: Knowing an LLM hallucinated after the user read it is already too late.
3.Runaway Multi-Model Costs & PII Leakage: Tracking exact token spend across 40+ rapidly changing models while keeping sensitive credentials and user data safe before transmission is a nightmare.
We built Observyze as an active reliability & observability layer rather than just a passive dashboard.
⚙️ What’s Under the Hood:
Non-Blocking <10ms Telemetry Ingestion: Fastify + Redis write buffer decouples trace capture from your agent's execution loop.
In-Memory Pre-TLS PII Scrubbing: 11 vectorized regex rules strip emails, credit cards, SSNs, and API keys locally in memory before payloads ever hit the wire.
Real-Time Circuit Breakers: Evaluates hallucination and safety thresholds inline to intercept bad outputs before they reach production users.
Exact 40+ Model Token Attribution: Real-time dynamic pricing tables with longest-prefix matching for OpenAI, Claude 3.5, Gemini, DeepSeek, and open-source models.
Zero-Code Drop-In Proxy: Swap `baseURL` in 2 lines of code using your existing OpenAI/Anthropic SDKs.
Our Early Access Gift to the Community:
To celebrate our Early Access launch, everyone who signs up gets a complimentary 90-Day Pro Tier pass (100,000 monthly traces, full evaluation suite, and real-time guardrails) with no credit card required
We'd love for you to try it in your agent workflows, kick the tires, and share your unfiltered feedback! I'll be here all-day answering questions in the comments. 🚀