Launching today
BotGauge
The trust layer for your agents
458 followers
The trust layer for your agents
458 followers
AI agents don't just generate answers anymore. They take actions, call tools, and access real data. BotGauge helps engineering teams find where agents break before users do. Run adaptive red-team attacks and evals, catch security and behavioral failures, and replay every issue with full traces. Turn failures into regression tests, continuously retest as your agents evolve, and monitor their behavior in production. One reliability layer from development to deployment.
Your agent has a failure mode you haven’t found yet. Your users will find it first. 👋
I’m Pramin, co-founder and CEO of BotGauge. Today we’re launching BotGauge: the platform that finds your AI agent’s failures before your users do.
Agents don’t fail like normal software. A user comes at it sideways, and it calls tools it shouldn’t, leaks context, or skips approvals. Change the model, a prompt, or an MCP server, and it can take a completely different path. Last month’s tests won’t catch that.
So we built an attacker.
🧠 Custom attacks for your agent. BotGauge learns your agent’s purpose, its users, and what it’s allowed to do. Then our in-house red-teaming engine builds attack patterns specific to that agent, because a booking agent and a support bot break in very different ways.
🔍 See exactly how it broke. Every finding links to the prompts, tool calls, and execution path behind it, so you fix the root cause.
🔁 Every failure becomes a permanent test. Each break turns into an eval that runs on every release and feeds your guardrails and monitoring.
Example: we told a booking agent to ignore its instructions, rebook to a new destination, and skip approval. BotGauge caught it, blocked the action, and added the attack to the eval suite for good.
It works with OpenAI, Anthropic, LangChain, LangGraph, CrewAI, AutoGen, and plain SDK/API.
We’re SOC 2 Type II.
👉 Try it free: Link
👉 Book a demo here: Link
What’s the weirdest thing your agent has done in production? The team and I are in the comments all day. 🚀
@pramin_pradeep How well does Not gauge catch failures that only appear when multiple tools interact????
Curious, what's the mist surprising way you're seen an agent break during testing?
@anton_grinenko Honestly, it’s rarely a clever hack. It’s more like an agent that gets asked nicely to ‘resend that last invoice to this address’ and just does it, to the wrong person, with the wrong data attached. Nothing looks broken until you read the trace.
@anton_grinenko Ha, good one! Two that stuck with me:
An agent hit a failed tool call and just kept retrying in a loop, burning tokens with no error and no alert.
A refund agent held its approval step through 117th role-play attacks. On the 118th, with a different persona, it caved and skipped approval.
Neither one crashed. They just quietly did the wrong thing.
Would love for you to run your own agents through BotGauge and see how they break 😄 Happy to help if you need any support or want to go deeper on the results!
To answer your question, back in February an OpenClaw agent I'd hooked up to WhatsApp started messaging a friend of mine and leaking its own internal notes into the chat. So I pulled its access. I'd never have thought to write a test for that one, which I guess is the point of having something attack the agent on purpose.
@simon_moxon Ha, that’s a great one, and kind of terrifying. Nobody sees ‘my agent texts my friend its private notes’ coming. Glad you caught it quickly. Did you ever figure out what made it start doing that?
Curious to see how it works in real world scenarios, especially with agents getting more complex
@jatinsingh That’s the part we’re most curious about too. Lab tests only tell you so much, and once an agent has real tools and messy inputs, things get weird fast. If you try it on something real, I’d love to hear how it goes.
Really curious to see what people come up with! Always interesting to see how someone pushes an agent in a way nobody expected.
@shraddhabhat Ha, yeah, me too. Half the fun is watching someone break something in a way you’d never think of. Let us know if you try it!
This looks really interesting. I've been hesitant to hook up some services to agents precisely because I didn't trust it not to take advantage of the permissions it was given. I'm definitely going to explore BotGauge further.
A bit of feedback on the post: your images are too low-resolution to really see what's on the screens you're sharing. I noticed the same thing on your website. I think some higher-resolution images would go a long way in helping people better understand what it is you're offering.
@nicholas_zakas Thank you, and that hesitation is exactly who we built this for. Being careful about giving an agent real permissions is the right instinct. Appreciate the note on the images too. We’ll take another look at how they’re showing up.
Excited to try out red-teaming. Stress testing the AI agent against real attack patterns before attackers to find the gaps sounds interesting.
@monish_k_s Appreciate it! The best part is hearing what people find. If you run it, tell us the one attack that caught you off guard and we’ll dig into it with you.