Founder of @openlit_io | Cyber security using AI @transilienceai https://xitter.cf/t.co/cz5Dr0pbHj
Noida
Joined August 2014
- Tweets90
- Following175
- Followers60
- Likes245
New @openlit_io demo: support-desk agent.
Scan the repo → fix unsafe tools → read traces + AI Analysis → stop duplicate tool calls.
Run cost: $0.0052 → $0.0002
Repo: github.com/openlit/openlit
youtu.be/mSOuJxW8cHs
If you’re building agents and want @openlit_io in the loop, DM me or email [email protected]
Feature idea → same, or github.com/openlit/openlit/i…
Collab → connect and we’ll work something out
Enterprise beta / trial → reach out, we’re taking users
Aman Agarwal retweeted
CVE-2026-85046 Google Chrome V8 Type Confusion Zero-Day 🚨
A webpage visit can be enough to exploit CVE-2026-85046.
The actively exploited V8 type confusion flaw enables sandboxed code execution in Chrome and affects the broader Chromium ecosystem.
Patch to 152.0.7977.82/.83+, restart, then hunt for suspicious browser-spawned processes and outbound connections.
Aman Agarwal retweeted
Your stack already has a home for telemetry.
OpenLIT datasource connectors let you bring it in:
ClickHouse · Tempo · Loki · Prometheus · Jaeger
Per-environment signal routing for traces, logs, and metrics.
Docs: docs.openlit.io/latest/openl…
OpenPlait: npmjs.com/settings/openplait…
Your stack already has a home for telemetry.
OpenLIT datasource connectors let you bring it in:
ClickHouse · Tempo · Loki · Prometheus · Jaeger
Per-environment signal routing for traces, logs, and metrics.
Docs: docs.openlit.io/latest/openl…
OpenPlait: npmjs.com/settings/openplait…
GPU utilization is a vanity metric for LLM inference.
nvidia-smi and DCGM tell you the SM was busy. Not KV cache OOM, throttle, or zombie VRAM. What to SLO instead:
openlit.io/blogs/gpu-util-is…
Aman Agarwal retweeted
Most #LLM #observability guides stop at a single chat completion. But production AI systems are far more complex.
This post is a joint effort by the OpenLIT and VictoriaMetrics teams. @openlit_io brings the #OpenTelemetry-native instrumentation for LLMs and agents, while VictoriaMetrics provides the storage and query layer for the resulting metrics, traces, and logs. We wrote it together to show how the two projects fit into a single, self-hosted observability pipeline, and to share the patterns and queries that worked well for us along the way.
Learn more 👇
hubs.ly/Q04nFzRY0
#Otel #OpenLIT #SDK VictoriaMetrics
1/6 Monitoring AI agents is painful. Infinite loops, high latency, and hard-to-track token costs make production deployments high-risk.
openlit.io/blogs/talk-to-you…
Say hello to Otter: OpenLIT's AI telemetry copilot designed to debug and optimize your AI apps conversationally. 👇
⭐ Star us on GitHub and support the standard for AI reliability:
github.com/openlit/openlit
6/6 Best part? OpenLIT is OpenTelemetry-native and self-hosted. Otter runs entirely within your private network—no data leaks.
5/6 🔒 Private VPC Compliance: Since OpenLIT is self-hosted, Otter analyzes telemetry logs locally, meeting HIPAA/GDPR constraints without leaking data to external APIs.
4/6 💸 Custom Pricing: Register token pricing for local models (Ollama/vLLM) so local cluster costs show up accurately alongside public cloud API costs.
3/6 🛡️ Query by Attribute: Ask Otter to retrieve traces matching any attribute you capture (session ID, custom safety tags) and it writes the ClickHouse query for you—then helps you draft a Rule Engine condition from what it finds.
2/6 🤖 Trace Analysis: Otter's routing/execution-path review flags repetitive tool calls in LangGraph/CrewAI traces and explains what likely caused them.
I wrote about a hard truth in RAG:
👉 Retrieval doesn’t eliminate hallucinations.
Even with good context, models can still give wrong answers.
In this blog:
* Where hallucinations come from
* Why retrieval isn’t enough (Dont trust the model!)
🔗 openlit.io/blogs/how-to-dete…
Been working with LLM systems lately and one thing stands out:
We track latency, tokens, cost… but barely look at the GPU.
which is where all the work happens.
Most debugging pain I’ve hit comes from this gap.
I wrote this blog to connect the dots:
openlit.io/blogs/gpu-monitor…
Aman Agarwal retweeted
We wrote a quick guide on adding observability to LLM apps with OpenTelemetry.
Trace prompts, debug issues, and track latency/cost in minutes 👇
openlit.io/blogs/how-to-add-…
